Data-Driven purchasing strategies: Price prediction models and strategy development Seray Mirasçı a,*, Aslı Aksoy b a LUT, School of Business and Management, Yliopistonkatu 34, 53850 Lappeenranta, Finland b Bursa Uludag University, Industrial Engineering Department, Gorukle Campus, 16059 Bursa, Turkiye A R T I C L E I N F O Keywords: Decision Making Machine Learning Strategic Purchasing Purchasing Management A B S T R A C T This study aims to provide price predictions for automotive steel materials using machine learning (ML) methods to develop purchasing strategies with business theories. The primary objective is to respond to the understanding of purchasing dynamics, emphasizing the importance of adopting an agile approach to support the development of purchasing strategies within organizations. The data set used in the study was provided by a global original equipment manufacturer (OEM) company. Clustering analysis is performed to determine the most strategic project cluster, and price prediction models are developed for the most strategic project cluster using ML methods. Artificial neural networks (ANN), and tree-based models (decision trees (DT), bagging, and boosting methods) are used for price prediction models. According to the model’s results, strategic purchasing suggestions are enhanced by incorporating a dynamic capability view (DCV) and information processing theory (IPT) to ensure adaptability and competitiveness in ever-changing purchasing dynamics. It was found that ANN per formed the best despite its black-box nature. While tree-based models did not perform as well as ANN, they provided valuable insight into the importance of different criteria weights in price prediction. Integrating advanced ML techniques like ANN and tree-based models significantly improved price prediction accuracy. ANN parameters were carefully optimized, and decision tree structures allowed for make quick price predictions. Additionally, incorporating business theories such as DCV and IPT. The research enhances strategic purchasing recommendations, ensuring adaptability and competitiveness amidst evolving purchasing dynamics. These findings contribute to streamlining purchasing processes and emphasize the transformative potential of inte grating business theories with ML methodologies in refining real-world analyses with precision. 1. Introduction In today’s changing world, it is essential to act with agility and strategic expertise, especially in supply chains (Aslam et al., 2018). The purchasing and supply chain functions play a crucial role in managing the supply of critical materials that are essential to an organization’s operations but have a high risk of supply chain disruptions (Liu et al., 2021; Sauer and Seuring, 2019). Understanding supply chain risk, defined as the probability of events arising from individual supplier is sues or market conditions potentially affecting the purchasing firm’s capability to meet customer demand or save customer safety, enables supply management professionals to implement improved management strategies (Zsidisin, 2003a, 2003b). The automotive industry, a significant global sector, holds immense economic influence and impact, necessitating effective management to address challenges and ensure overall well-being (Masoumi et al., 2019). The purchasing departments of this industry, along with their suppliers and partners, generate a vast amount of data within the organization’s network, and this holds great potential for creating additional value, but it is frequently not fully exploited or realized (Allal-Chérif et al., 2021). Proactively managing risks and collaborating with suppliers empower purchasing and supply chain functions to ensure a reliable flow of crit ical materials (Jaipuria et al., 2015; Liu et al., 2021). Machine learning (ML) has revolutionized predictive analytics, especially in price fore casting, by effectively identifying complex data patterns and enabling real-time predictions. Techniques like Artificial Neural Networks (ANN) and tree-based models are widely used for their accuracy in price pre dictions across diverse sectors, including supply chain management (Chen and He, 2019; Gu et al., 2021). Furthermore, strategic risk miti gation strategies such as price predictions, supplier diversity, demand, * Corresponding author. E-mail addresses: seray.mirasci@lut.fi (S. Mirasçı), asliaksoy@uludag.edu.tr (A. Aksoy). Contents lists available at ScienceDirect Expert Systems With Applications journal homepage: www.elsevier.com/locate/eswa https://doi.org/10.1016/j.eswa.2024.125986 Received 11 March 2024; Received in revised form 16 November 2024; Accepted 28 November 2024 Expert Systems With Applications 266 (2025) 125986 Available online 5 December 2024 0957-4174/© 2024 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ). https://orcid.org/0000-0002-2971-2701 https://orcid.org/0000-0002-2971-2701 mailto:seray.mirasci@lut.fi mailto:asliaksoy@uludag.edu.tr www.sciencedirect.com/science/journal/09574174 https://www.elsevier.com/locate/eswa https://doi.org/10.1016/j.eswa.2024.125986 https://doi.org/10.1016/j.eswa.2024.125986 http://crossmark.crossref.org/dialog/?doi=10.1016/j.eswa.2024.125986&domain=pdf http://creativecommons.org/licenses/by/4.0/ and inventory management help minimize the impact of critical mate rials. With a proactive approach, managers can minimize the impact of supply chain disruptions on the availability of critical materials (Paul et al., 2023). As one of the lessons learned, the pandemic has highlighted the risks of relying on a single supplier for critical materials and underscores the fact that the purchasing and supply chain departments should consider diversifying their supplier base to reduce the risk of supply disruptions. This study aims to propose the development of a pricing prediction model for purchasing steel materials that aligns with information pro cessing theory (IPT) and dynamic capability view (DCV). This model aims to support profitability strategies quickly and easily, addressing challenges such as time constraints and reliance on human experience. By leveraging the framework, this initiative aims to streamline pur chasing strategies, ensuring alignment with the company’s goals of digital transformation and risk management capability. Effective man agement of purchasing processes of steel materials in an original equipment manufacturer (OEM) company is the basic problem for this study. This study aims to provide accurate price predictions for automotive industry purchasing departments based on project and material data. These predictions enable buyers to efficiently compare and interpret supplier quotations, optimizing purchasing decisions and reducing financial risks. In accurate price predictions can lead to erroneous budgeting, impacting a company’s profitability and hindering pur chasing strategy development and negotiation with suppliers. Accurate price anticipation empowers buyers in negotiations by understanding supplier constraints. In buyer price negotiations, the current manual process lacks the efficiency of technology-driven data preparation tools, potentially leading to errors. The overall goal is to develop robust pur chasing strategies for steel materials, ultimately improving efficiency and cost-effectiveness. Consequently, the crucial point of this study is to integrate price predictions into business theories that fit with purchasing departments. The discussions underscore that the convergence of pre dictive pricing and strategic planning constitutes a pivotal concern in practical operations, necessitating meticulous attention to the imple mentation of technology-driven controls to reduce potential errors and improve decision-making processes. This study introduces a dual-faceted contribution to the field, addressing a critical research gap by uniquely integrating ML methods with business theories. Although existing computer science literature explores various ML techniques for price prediction across different material and sector domains (Kureljusic and Karger, 2024; Hammann, 2023; Khan, 2021; Arslankaya and Toprak, 2021; Chen and He, 2019; Sakız and Gencer, 2018; Čeperić et al., 2017; Liu et al., 2017; Bayram et al., 2016; Yan et al., 2006), business literature investigates the role of business theories in advancing purchasing and supply chain functions (Akpinar and Vincze, 2016; Hendry et al., 2019; Chirumalla, 2021). Despite the growing body of work on ML in price prediction, limited amount of study (Kim et al., 2016) address how business theories, as DCV and IPT can be integrated to enhance strategic decision-making within procurement processes in a context of computer sciences. There are still a significant void remains at the intersection of these domains. Qin et al. (2024) highlighted that working at the intersection of business and computer science holds promising potential for advancing intelli gent decision-making, social governance, capital management, Industry 4.0, and innovation. This study aims to bridge this gap by incorporating both ML models and business theories to provide a comprehensive framework for real-time, data-driven purchasing strategies, which has not been deeply explored in existing literature. The research question of this study is as follows: “How can the integration of ML models and business theories such as DCV and IPT be integrated to enhance price prediction accuracy and strategic decision-making in purchasing?”. The remainder of this paper is organized as follows. Section 2 presents a theoretical background. Section 3 explains the methods. Findings are provided in Section 4. Discussions are presented in Section 5, and conclusions are presented in Section 6. 2. Theoretical background With the rapid development of information technologies in recent years, the increase in the volume of data generated by supply chain partners has made data management and its applications an important tool to improve supply chain performance. The emphasis on the signif icance of data analytics in business and management is directed towards addressing concerns like long-term contracts, material supply levels, and spot market purchasing. This facilitates buyers in making well-informed purchasing decisions (Zhang et al., 2023). Data management, involving the application of various statistical and ML methods to datasets, embodies a novel facet of contemporary business intelligence, facilitating the extraction of valuable insights, improving decision-making processes, and providing competitive ad vantages to organizations operating within today’s data-centric (Moskowitz et al., 2011). Purchasing departments leverage price pre diction methods to dynamically respond to evolving business conditions, thereby optimizing supply chain operations, and effectively managing budgets. This innovative approach entails a three-stage optimization model that incorporates circular economy principles, aiming to tackle uncertainties while addressing environmental and social considerations inherent in contemporary purchasing and strategic management prac tices (Foroozesh et al., 2023; Kocabıyıkoğlu et al., 2016). ML, which belongs to the field of artificial intelligence, deals with the development of algorithms that can be learned from data. ML methods accelerate and streamline decision-making by providing useful business insights (Raviteja et al., 2022). Various ML algorithms, like neural net works, decision trees, support vector machines, and others, are used for prediction and classification tasks (Blumberg and Thompson, 2022; Kachamas et al., 2019; Sharma and Soni, 2020). ANNs arise from the mathematical modelling of the learning ability in the human brain and are learning systems that learn through exam ples. These neural networks are composed of artificial neurons. Each neuron has a weight, and the information stored in the system is stored in these weight values (Öztemel, 2012). ANNs are parallel computing systems using processors with several interconnections, developed by utilizing the working principle of biological neural networks based on the fact that the human brain is better at perception problems (Jain et al., 1996). The ANN algorithm is used to discover non-linear, complex relationships between input and output variables (Truong et al., 2021). The most widely used form of ANN for predicting is feed-forward Multi- Layer Perceptron (MLP) to approximate nonlinear functions (Fajar and Nurfalah, 2021). Several studies have employed various approaches such as ANN to develop predictive models for diverse applications, including cost prediction for sheet metal parts (Verlinden et al., 2008) and stamping dies (Özcan and Fığlalı, 2014), sales price predictions (Kuo and Xue, 1998), passenger air transport demand during the Covid- 19 pandemic (Li et al., 2021), as well as copper price prediction (García and Kristjanpoller, 2019; Khoshalan et al., 2021). Decision tree (DT), proposed by Quinlan (1986), is a supervised learning model and can be used for regression and classification prob lems based on continuous or discrete data (Wu et al., 2022). Tree-based models generate simple and efficient solutions for multiple variables (Krishna et al., 2023). According to literature research, ML-based price prediction models are widely used in industrial applications. Price prediction of critical materials is vital for companies to enhance profitability, but under standing how to use predicted prices is equally crucial. This study is focused on the integration of price-prediction models to develop pur chasing strategies in an OEM company for steel materials. 2.1. Dynamic capability view View in supply chain risk management highlights the need for S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 2 organizations to continuously develop and leverage dynamic capabil ities for effective risk identification and adaptation. For example, the COVID-19 pandemic underscored the central role of dynamic capabil ities, as rapid response and reconfiguration of resources proved critical to improving supply chain resilience (Kähkönen et al., 2023). DCV, in troduces a multidimensional measurement instrument for supply chain resilience, with potential predictive value for supply chain performance and operational vulnerability (Chowdhury and Quaddus, 2017). Utiliz ing the DCV shows that uncertainty motivates firms to integrate business strategies and develop strategic flexibility, resulting in enhanced supply chain coordination and improved operational performance, with supply chain coordination mediating the link between strategic flexibility and operational performance (Rehman and Jajja, 2023). Akpinar and Vincze (2016) used the DCV view to assess the impact of risks, resource management, and supplier positions. They examined how power differences affect competition levels and how common stakes influence cooperation levels in the context of changing competition dynamics between Volkswagen Group and Porsche AG from 2001 to 2012, with a focus on threats. Hendry et al. (2019) analyzed 14 UK food sector firms using the dynamic capabilities framework, examining their response in the sensing, seizing, and transforming stages. The DCV can support price prediction model results and strategy development by providing a framework for organizations to adapt and respond effec tively to changing market conditions, including pricing fluctuations (Shaoming et al., 2015). It enables organizations to continuously adjust their purchasing strategies (Jones and Knoppen, 2018), enhancing supply chain resilience and performance (Kähkönen et. al., 2023). DCV supports the reduction of costs for firms and the improvement of com pany performance (Zott, 2003). The cost adaptability and effectiveness align with the strategic flex ibility components, sensing, seizing, and reconfiguring capabilities of DCV, which leads to improved supply chain coordination and opera tional performance (Rehman and Jajja, 2022). The DCV enables orga nizations to incorporate predictive models into their purchasing strategies, promoting agility and competitiveness in pricing decisions and supplier relationships (Bahrami and Shokouhyar, 2022; Susitha et al., 2024). 2.2. Information processing theory The integration of ML methodologies within the procurement and pricing frameworks of the automotive industry can be contextualized within the theoretical paradigm of IPT. This theoretical framework, foundational within cognitive psychology, elucidates the mechanisms by which humans acquire, retain, and retrieve information, akin to computational systems (Galbraith, 1974). ML algorithms, functioning analogously to cognitive processes, engage in the analysis of extensive datasets to discern salient patterns and correlations. IPT consists of two components: information processing requirement reduction and infor mation processing (Li et al., 2021). This process mirrors the cognitive operations inherent in human decision-making, wherein perceptual in puts undergo cognitive processing to inform subsequent actions. ML models reflect IPT by emulating human memory processes: they encode, store, and retrieve learned patterns to support decision-making (Liu, 2022), much like how IPT describes human cognition as a sequence of encoding, storing, and recalling information to solve problems. Additionally, these algorithms exhibit selective attention mecha nisms, akin to human cognitive functions, which facilitate the identifi cation of relevant information amidst noisy data. Furthermore, the iterative learning process inherent in ML algorithms, facilitated by feedback mechanisms, mirrors human adaptive learning, resulting in the refinement and enhancement of predictive capabilities over time. Thus, the deployment of ML techniques in automotive purchasing and pricing endeavours represents a computational emulation of human cognitive processes, thereby augmenting decision-making (Melville and Ramirez, 2008) within the multifaceted landscape of supply chain management. Existing research highlights a gap in the integration of ML techniques with strategic business theories in procurement and supply chain con texts (Qin et al., 2024). While numerous studies focus on ML applica tions for price prediction (Verlinden et al., 2008; Özcan and Fığlalı, 2014; Kuo and Xue, 1998), few address how these models can be effectively combined with theoretical frameworks such as the DCV and IPT. This study aims to bridge this gap by employing advanced ML techniques, specifically ANN and tree-based models, for price prediction while embedding them within business theories as DCV and IPT frameworks. This approach not only enhances predictive accuracy but also provides a strategic lens for understanding and improving decision- making in procurement, addressing the challenges of complexity and uncertainty in dynamic market environments. 3. Methods This study aims to formulate precise purchasing strategies and price prediction models for an OEM in the automotive industry. The study consists of several key phases. It begins with data preparation, which involves the identification and removal of noisy data points in the dataset. This is followed by project clustering, where the optimal num ber of clusters is determined, and cost-based product groups and clusters are generated. The study then proceeds to price predictions using both ANN and tree-based algorithms. A comparative evaluation of these prediction models is provided, and finally, purchasing strategies are formulated based on the price predictions. The framework of the study is summarized in Fig. 1. 3.1. Data preparation It’s crucial to emphasize the vital importance of setting reasonable prices for steel materials, yet pricing in many enterprises lacks robust technological oversight. Manual errors by buyers can lead to significant price deviations, including meager, high, or even negative prices. Therefore, technological controls are necessary to avoid operational challenges. The dataset of 732 unique steel materials used in the study includes steel materials features such as project number, consumption, width, length, cutting type, thickness, quality, product group and prices. Each steel material is associated with a project, often linked to a specific car model. However, some car models are no longer in mass production, and their sub-components are solely available as spare parts, resulting in lower annual demand. This leads to instances of steel materials lacking project information due to the absence of mass production demand. The project number is known if the project is in mass production. However, the dataset also includes some projects that are not in mass production and are only manufactured when spare parts are needed. The presence of noisy data is a common problem that can lead to incorrect analysis of the dataset, resulting in incorrect recommendations. Purchasing datasets can be prone to noisy data from several sources, such as price entry errors, out-of-date files, and incorrect calculations. 3.2. Clustering methods The K-means method aims to reach simple and understandable data sets from large and complex data (Patel et al., 2022). In the K-means method, during the initialization phase, each data point is assigned to the nearest random cluster center. In the iteration phase, the corre sponding cluster center is updated to be the average of the data points, and the distance of the respective data point to the cluster center is calculated using the selected distance metric to determine the error value. In the refinement phase, each data point is reassigned to its nearest cluster center again. The stopping process occurs when no data point can move between clusters. Determining the optimal number of clusters is a challenging problem in data analysis. The Connectivity index, Dunn index, and Silhouette S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 3 index are used to evaluate the performance of the clustering model. The Connectivity index observes local densities in the data and separates data groups into clusters according to their closest neighbors, the Silhouette index indicates the stability of objects in clusters according to their locations, and the Dunn index is the ratio of the shortest distance and longest distance in the same cluster (Shobha and Asha, 2017). 3.3. Price prediction methods In this study, price forecasting models were developed using ANN and tree-based methods. ANN is a powerful modelling tool for complex input–output re lationships, ANN structure comprises interconnected neuron clusters, featuring three layers: the input layer for training, the output layer for results, and adaptable hidden layer(s) bridging the input and output for problem-specific adjustments (Maghraoui et al., 2022; Bourdeau et al., 2019). Network training involves adjusting neuron weights, initially set randomly. Various training algorithms are available for ANNs, with the widely used Levenberg-Marquardt (LM) algorithm (Ranganathan, 2004). The LM algorithm is a training algorithm that uses the combi nation of the gradient descent method and the Gauss-Newton method known as the minimization algorithm (Lourakis, 2005). Small damping parameter values trigger Gauss-Newton updates, while large values prompt gradient descent updates. LM training algorithm uses Eq. (1) to update the weights of the ANN model: wk+1 = wk − ( JT k Jk + μI )− 1JT k ek (1) where J is the Jacobian matrix, w is the weight vector, e is the error vector, I is the identity matrix and µ is the combination coefficient. In Eq. (1), the LM algorithm is converted to the Gauss-Newton iteration method for smaller values of the combination coefficient µ and for larger values of µ iterative direction of the LM algorithm is similar to the gradient descent method (Mukherjee and Routroy, 2012). Bayesian Regularization (BR) training algorithm first introduced by MacKay (1992) minimizes a combination of squared errors and weights and then determines the correct combination so that the network has good generalization performance at the end of training (Fiorentini et al., 2023). In the BR algorithm the performance function of the model is replaced by F(y,w) = α*Es + β*Ew (2) where Ew is the sum of squared network weights and, Es is the sum of squared network errors, α and β are objective parameters to be estimated during training (Zhao et al., 2011). The BR algorithm updates the pos terior distribution of the ANN weights using Bayes’ rule and maximizing the posterior probability of w is equivalent to minimizing the regularized objective function in Eq. (2) (Kayri, 2016). The scaled conjugate gradient (SCG) algorithm adjusts the weights of neurons according to the SCG method, which offers faster convergence than the steepest descent methods (Møller, 1993). DT is a kind of non-parametric supervised learning algorithm, per forms a greedy search to determine the optimal splitting points in a tree. The regression tree model recursively divides each region into two sub- regions based on the input parameter(s) of the training data set and specifies the corresponding output value for each sub-region (Cui and Fearn, 2017). The main advantages of DTs are that the results are easy to interpret, and they perform feature selection indirectly, as the top nodes of the tree are the most important variables in the dataset. The main limitation of DT is that as a DT grows and becomes too complex, high variance and low bias can be observed, leading to overfitting (Kaparthi and Bumblauskas, 2020). Ensemble methods, which are designed to reduce variance and bias, are classified into bagging-based and boosting-based methods (Ruiz de Miras et al., 2023). Ensemble learning algorithms are designed to train low-accuracy models (known as weak learners) and combine the pre dictions of weak learners to generate a high-accuracy model (known as strong learners) (Al-Qudah et al., 2022). DTs are trained using a bagging mechanism and ensemble in an RF algorithm (Polikar, 2012). RF is the most used bagging-based ML algorithm (Paturi and Cheruku, 2021, Ruiz de Miras et al., 2023) and one of the most successful ML methods Fig. 1. The framework of the study. S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 4 (Breiman, 2001). RF uses classification trees as weak learners, and the number of trees and the size of the random subset of the features to consider at each split are the most important hyperparameters of an RF (Ruiz de Miras et al., 2023). Since DTs in RF grow until all leaves are pure, by setting a maximum depth or requiring a minimum number of samples per node before or after splitting, the growth of the tree can be limited (Bentéjac et al., 2021). Boosting algorithms iteratively create multiple models by using a weak learner and focus on reducing bias. The boosting learning algorithm combines the prediction of several weak or medium predictors to improve the robustness of a single estimator (Al- Qudah et al., 2022). In gradient boosting, the ensemble model consists of a weighted sum of weak learners (Ruiz de Miras et al., 2023). Gradient boosting algorithms are used to ensemble tree models and provide outstanding classification performance in labelled data analysis (Hung, 2022). Gradient boosting algorithms are designed to approximate the function F*(x), assigning instances x to their output y, by minimizing the expected value of a given loss function L(y, F(x)). Gradient boosting provides an additive approximation to F*(x) as a weighted sum of functions shown in Eq. (3): Fm(x) = Fm− 1(x)+ ρmhm(x) (3) where ρm is the weight of function hm (Bentéjac et al., 2021). Eqs. (4)–(8) are used to evaluate the performance of trained ML models based on the coefficient of determination (R2), the mean abso lute error (MAE), mean square error (MSE), mean absolute percentage error (MAPE), Nash-Sutcliffe efficiency (NSE): R2 = 1 − ∑N i=1(yi − ŷi) 2 ∑N i=1(yi − yi) 2 (4) MAE = 1 N ∑N i=1 |ŷi − yi| (5) MSE = 1 N ∑n i=1 (ŷi − yi) 2 (6) MAPE = 1 N ∑N i=1 ⃒ ⃒ ⃒ ⃒ ŷi − yi yi ⃒ ⃒ ⃒ ⃒ (7) NSE = 1 − ∑N i=1(ŷi − yi) 2 ∑N i=1(yi − yi) 2 (8) where yi is the actual value, ŷi is the predicted value, yi is the mean value and N is the total number of data. R2 is the proportion of total variability in the dependent variable that is explained by the indepen dent variables. and a. The R2 and NSE values exceeding 0.40 signify an acceptable degree of model prediction performance and higher R2 and NSE values indicate a greater degree of model fit, whereas lower values of MAE, MSE and MAPE correspond to more accurate predictions. (Citakoglu, 2017; Bayram and Citakoglu, 2023). To compare the performance of the developed models statistically, the results of the Kruskal-Wallis test are evaluated. The Kruskal-Wallis test, proposed in 1952, is a non-parametric method for testing the hy pothesis that independent groups are from the same distribution (Afrasiabi et al., 2021). The test does not assume a normal data distri bution and was conducted to ascertain the validity of the null hypoth esis, which states that the k groups originate from the same population and possess the same median value (Devi Priya et al., 2022). This was achieved by using the H statistics given in Eq. (9): H = [ 12 n(n + 1) ∑k j=1 T2 j nj ] − 3(n + 1) (9) where n is the total sample size, k is the number of groups, Tj is the sum of ranks for groupi, and nj is the sample size of groupi. As demonstrated by Kruskal (1952), the statistic exhibits a limiting null distribution that can be approximated by a chi-squared distribution with k-1 degrees of freedom (Spurrier, 2003). Taylor diagrams presented by Taylor (2001) used to provide a visual representation to compare the performance of developed models (Demirbay et al., 2022). The Taylor diagram offers a methodology for the plotting of three statistical values on a two-dimensional graph, which indicate the degree of correspondence between observed and predicted patterns. These statistics facilitate the determination of the extent to which the overall root mean square difference in patterns is attributable to variance differences and the extent to which it is attrib utable to poor pattern correlation Taylor (2001). To facilitate the com parison of various ML methods, the Taylor diagram used. This diagram shows the compatibility between the predictions made by different models and the actual data (Bayram and Çıtakoğlu, 2023; Demir, 2022), thereby providing insights into their respective compatibilities and performances in relation to real-world scenarios. 3.4. Analytical approaches in Developing purchasing strategy In today’s dynamic procurement landscape, the adoption of advanced analytical methods like ML is crucial for informed decision- making. To develop accurate purchasing strategies, it is vital to bridge ML insights with business theoretical frameworks like the DCV and IPT. This integration enhances adaptive capabilities and streamlines infor mation processing, empowering organizations to navigate procurement complexities with agility and precision. 4. Findings There are several purchasing-based problems faced by the OEM buyer. The price data set used in this study consists of data from different projects. There are 732 rows covering 18 different projects in the dataset. To determine effective purchasing strategies, the projects are classified and to determine the true prices of steel materials products for purchasing strategies, price prediction models are developed with different ML algorithms. The R Studio software package, version 1.4.1106 was used for the clustering, and MATLAB R-2022b was employed for the development of ML models. The models were devel oped on a 1.50-GHz Intel Core i5 processor with 8 GB of memory. The existing dataset was analyzed graphically, revealing noise in the form of zero and negative values. An R script was then used to filter out these instances, resulting in a reduction of rows from 732 to 414. Additionally, the project has been clustered for improved organization and analysis. The K-means clustering algorithm is highly regarded for its flexi bility, efficiency, and ease of implementation (Ikotun et al., 2023), which has led to the K-means algorithm being identified as one of the most significant clustering methods in data analysis (Ismkhan, 2018; Wu et al., 2008). In order to identify project clusters of strategic importance, the K-means algorithm is employed. The application of the K-means algorithm for different numbers of clusters is illustrated in Fig. 2. The optimal number of clusters is also validated by connectivity, Silhouette and Dunn index values for different numbers of clusters, as shown in Table 1. Table 1 shows the clustering validation indices for Connectivity, Dunn and Silhouette method. The maximum value of the Dunn index is associated with the optimal number of clusters (Saha and Bandyo padhyay, 2012), and the Dunn index is maximized with four clusters. The Connectivity index is associated with minimizing the maximum diameter between all the clusters, which in turn aims to minimize the values, thus aligning with the optimal number of clusters (Saha and Bandyopadhyay, 2012). The minimum value of the Connectivity index is also observed with four clusters. It is crucial to validate and cross-check whether the defined four clusters represent the optimal number of clusters for this dataset by assessing the performance and quality of the S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 5 Fig. 2. Application results of K-means algorithm (a) for three clusters, (b) for four clusters, (c) for five clusters, (d) for six clusters, (e) for seven clusters. S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 6 selected clusters. Indicators such as Silhouette method (Ay et al., 2023; Brida et al., 2014) and Elbow method (Soubeiga et al., 2025; Bansal et al., 2024) can be used to evaluate the performance or quality of the clusters (Huang et al., 2024). A Silhouette index value approaching one is indicative of a data point that is well-matched to its assigned cluster (Ay et al., 2023). As illustrated in Table 1, the Silhouette Index value is most closely aligned with a value of one when employing a four-cluster solution, with a value of 0.6342. A graphical representation of the Silhouette method is provided in Fig. 3(a). As illustrated in Fig. 3(a), the Silhouette method suggests the number of clusters should be set at four. The graphical representation of the Elbow method, as illustrated in Fig. 3(b), enables the identification of the point at which the addition of further clusters no longer produces a notable impact (Bansal et al., 2024). The Elbow method identifies the optimal number of clusters by locating the elbow in the graph, where there is a sharp decline followed by smaller changes (Martino et al., 2023; Cui, 2020; Patel et al., 2022). This indicates that the addition of further clusters beyond this point has a minimal impact on the quality of the clustering. Fig. 3(b) illustrates the Elbow Method, demonstrating the point at which a distinct bend or elbow in the graph is observed. Following an evaluation of the optimal number of clusters using the Dunn Index, Connectivity Index, Silhouette Index, and the Elbow Method, all indicators consistently indicate that four clusters represent the optimal choice for this study. The results of the clustering algorithm concerning project numbers can be seen in Table 2. The columns represent the project numbers and the intersection cells represent the number of references for the selected project in the cluster. The classification of projects into four clusters is approved by the OEM, and Cluster_4 is identified as the most strategic cluster with high product diversity. Price prediction model has developed for the projects in Cluster_4. Following this classification, a price prediction model was specifically developed for projects within Cluster_4. This model utilized technical information from the most strategically clustered data points, incorporating independent variables such as consumption, length, width, cutting type, thickness, quality, and product group. The description of the independent variables is shown in Table 3. The dependent variable is the price of the steel materials product. Time se ries plot for the price label can be seen in Fig. 4. In this study, the DT, RF, GBT, and ANN algorithms are selected as the ML algorithms to predict the price of steel material products. The DT algorithm is one of the most widely used ML algorithms because the results are easy to interpret, the top nodes of the tree are the most important variables in the data set, and feature selection is performed simultaneously (Chen et al., 2003). Although the DT is known to be prone to overfitting, it has been described as a fast algorithm (Justo- Silva et al., 2021). Two ensemble ML algorithms (RF and GBT) are used because ensemble ML techniques usually make better predictions than a single learner (Li et al., 2023). ANN is believed to be the most commonly used ML algorithm (Liu et al., 2018). One of the drawbacks of the ANN models is that results are presented as black-box decisions (Kang et al., 2020). The ML algorithms are run with the following hyperparameters: maximal depth of 7 and minimum leaf size of 4 for the DT algorithm, minimum leaf size of 8, number of learners 30 for the RF algorithm, minimum leaf size of 8, number of learners 30, learning rate of 0.7 for the GBT algorithm. In the price prediction model based on ANN, a network structure with seven inputs and one output is developed. The LM, BR and SCG algorithms are used for training. The optimal ANN parameters were determined through various trials, resulting in a network with two hidden layers, six and four neurons in each hidden layer. A graphical summary of the trials conducted to determine the optimal structural configuration of network is provided in the Appen dices A, B, and C. The architecture of the ANN used in this study is shown in Fig. 5. The observed data set was split into three sets, training, test and validation, before running ML algorithms. Training and test sets are used for model training and validation set is used to evaluate trained model performance with unseen data. The dataset consists of 198 sample data, 30 of which are randomly selected for the validation set. The remaining 80 % of the data was used for training and the other 20 % for testing. 5- fold oss-validation method was used for the training data set of tree- based models. Table 1 Clustering Validation Indices. Cluster Size Connectivity Dunn Silhouette 3 12.7560 0.0252 0.6063 4 3.3778 0.0773 0.6342 5 15.9484 0.0398 0.6039 6 27.8802 0.0479 0.5762 7 34.0944 0.0479 0.6215 Optimal Score 3.3778 0.0773 0.6342 Optimal Cluster Size 4 Fig. 3. Graphical representation of the (a) Silhoutte method and (b) Elbow method. S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 7 The ML models were tested using a validation set that was not used for training. Table 4 presents the MSE, MAE, R-Square, NSE, MAPE performances of the models on the training and validation sets. Addi tionally, the CPU time required for training the developed ML models is presented in Table 4. Based on the performance values of the applied ML algorithms on the training dataset in Table 4, the RF model has the minimum MSE, MAE and CPU time, and the maximum R2 and NSE among the tree-based models. Additionally, the ANN with LM algorithm has the minimum MAE, MAPE, CPU time and maximum R2 and NSE values compared to other ML models. ANN models demonstrate superior performance in terms of CPU time when compared to tree-based methods, particularly when employing BR and LM algorithms. For the validation dataset, GBT, RF and ANN with LM models have better performance values. However, the ANN models outperform the tree-based models in all performance metrics for the validation set. A notable decline was observed in the R2 and NSE values between the training and validation sets for both the ANN and tree-based models. However, the ANN with the LM algorithm exhibited a more moderate decline, with R2 and NSE values decreasing by approximately 12.2 % and 24.3 %, respectively. In contrast, the RF algorithm demonstrated a more pronounced decline, with R2 and NSE values decreasing by approximately 28.7 % and 44.17 %, respectively. The Taylor diagrams were presented for both the training and vali dation data sets in Fig. 6. As illustrated in Fig. 6, Taylor diagrams reveal that the DT and GBT models tend to overestimate the variability of the actual price for the training set (Fig. 6a), while underestimating it for the validation set (Fig. 6b). Furthermore, DT and GBT models exhibit lower correlation coefficients and higher standard deviation values than other ML models. The ANN models demonstrate a closer alignment with the actual price (depicted by the solid red line) for both the training and validation sets, while the ANN with LM model represents an optimal equilibrium between standard deviation and correlation coefficient, with robust predictive performance. The Kruskal-Wallis test results of the price and developed ML models for training and validation data sets are presented in Table 5. Fig. 7 il lustrates the graphical representation of the Kruskal-Wallis test statistic. The Kruskal-Wallis test was performed at the 95 % confidence in terval with the critical value of p = 0.05. As indicated in Table 5, the p- Table 2 Clustering analysis results for four clusters. P1 P2 P3 P4 P5 P6 P7 P8 P9 P11 P12 P14 P16 P17 P18 Cluster_1 0 0 0 0 0 0 0 0 0 1 25 1 3 7 54 Cluster_2 0 0 0 0 0 0 0 0 0 0 0 0 2 1 22 Cluster_3 2 0 17 34 5 0 0 6 20 1 0 0 0 0 0 Cluster_4 1 1 19 149 16 2 1 2 22 0 0 0 0 0 0 Table 3 Independent variables. Description Type Consumption Annual consumption as tons Real numeric data [1,425–2516] Length Length of the material coil(mm) Real numeric data [540–2320] Width Width of the material coil (mm) Real numeric data [125,5–1460] Cutting Type The cutting type of material Integer numeric data [1–5] Thickness The thickness of the material (mm) Real numeric data [0,5–8] Quality Quality number of the material Integer numeric data [1–7] Product Group Number of the product group Integer numeric data [1–3] Fig. 4. Time series plot for the price label. Fig. 5. The architecture of ANN for price prediction model. S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 8 values of the developed ML models exceed 0.05, thereby rejecting the null hypothesis. This implies that there is no significant difference be tween the means of price value and the outputs of the developed ML models’ for both the training and validation sets, as demonstrated in Fig. 7. The Chi-squared values obtained from the Kruskal-Wallis test are used to quantify the extent to which the rankings of the outputs of the developed ML models differ from the price under the null hypothesis. A higher Chi-squared value indicates a greater discrepancy between the ranking of the outputs of the developed ML models and the price. As evidenced by the Kruskal-Wallis test results presented in Table 5, ANN- based models demonstrate superior performance compared to tree- based models. The ANN with LM model demonstrates the most optimal performance, as evidenced by the lowest Chi-squared value and the highest p-value for both the training and validation sets. Fig. 8 displays the fitted line graphics of the validation set for developed ML models. There are multiple outliers and a noticeable de viation from the fitted line for many data points in tree based ML models (in Fig. 8a, b, and c). The scatter points of ANN-based models (in Fig. 8d, e, and f) are closer to the fitted line compared to tree-based models, indicating that the ANN-based models capture the relationship between the predicted and actual data more accurately. ANN with LM model provides the best performance in terms of fitting the data for the vali dation set. These results show that the developed model can be used for un recognized products. Through the application of the tree-based models, the tree structure helps to identify the breakdown points of the input parameters for price prediction model. The buyer has determined that the consumption and the cutting type are the most important parameters and that any change in these parameters can have a significant impact on the price of the product. ANN model predicts a substantial price increase in steel materials due to market trends, dynamic capabilities allow the organization to respond by exploring alternative suppliers or negotiating more favorable terms. For example, the organization may quickly switch to a supplier Table 4 Performance values of ML algorithms for the training and validation set. DT GBT RF ANN with SCG ANN with BR ANN with LM ​ Training Set MSE 2025.126 2011.472 1613.243 2009.704 1819.964 2017.746 MAE 25.278 28.906 22.379 25.467 22.413 20.191 R-Squared 0.732 0.734 0.787 0.876 0.912 0.953 NSE 0.659 0.642 0.778 0.874 0.912 0.951 MAPE 0.018 0.010 0.012 0.010 0.009 0.004 CPU Time (sec) 3.200 1.983 1.421 1.559 1.204 1.199 ​ Validation Set MSE 5163.682 4469.348 4492.575 3061.996 2349.172 2225.078 MAE 42.287 33.632 39.361 28.698 26.843 22.677 R-Squared 0.450 0.471 0.560 0.665 0.757 0.836 NSE 0.350 0.437 0.435 0.615 0.704 0.720 MAPE 0.027 0.021 0.025 0.018 0.017 0.020 Fig. 6. Taylor diagrams (a) for training set (b) for) validation set. Table 5 The Kruskal-Wallis test results. (Chi- squared, p- value) Training Set DT GBT RF ANN with SCG ANN with BR ANN with LM Price (1.08, 0.298) (0.91, 0.339) (0.09, 0.764) (0.02, 0.884) (0.01, 0.924) (0, 0.987) ​ Validation Set Price (0.24, 0.624) (1.69, 0.194) (2.18, 0.139) (0.03, 0.858) (0.02, 0.882) (0.01, 0.919) S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 9 with a better price offer or initiate negotiations with the current supplier to secure a more cost-effective contract. The tree structure can identify critical factors influencing price changes. In response, dynamic capa bilities facilitate strategic portfolio management, potentially leading to the discontinuation of lower-quality materials and the prioritization of higher-quality options, contributing to enhanced portfolio agility. In parallel, the utilization of ANNs and tree-based models within the organization’s predictive modelling framework resonates with the principles of IPT. IPT provides insights into how humans process, encode, store, and retrieve information, akin to computational systems. By leveraging ANN and tree-based models, the organization effectively emulates cognitive processes, encoding, processing, and retrieving in formation to inform decision-making. This emulation mirrors the cognitive operations outlined in information processing theory, facili tating the organization’s ability to analyze vast datasets and derive actionable insights. Thus, the integration of ANN and tree-based models within the organization’s strategic framework not only enhances decision-making capabilities but also aligns with the theoretical un derpinnings of information processing, fostering a deeper understanding of cognitive mechanisms in the context of predictive analytics. Furthermore, dynamic capabilities enable the seamless integration of sustainability considerations into purchasing decisions, aligning with Fig. 7. Graphical representation of the Kruskal-Wallis test results (a) for training set, (b) for validation set. S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 10 corporate sustainability goals. The organization can leverage dynamic capabilities to evaluate and select suppliers based on their environ mental and social practices, ensuring that purchasing aligns with sus tainability objectives. This integration of dynamic capabilities with predictive models enhances the organization’s ability to adapt, inno vate, and optimize its purchasing strategies, ultimately improving its competitiveness within the industry. The use of dynamic capabilities in conjunction with ANN and tree-based models empowers the organiza tion to make informed, agile, and sustainable purchasing decisions, ul timately contributing to improved financial outcomes and a heightened competitive advantage. The integration of dynamic capabilities with price prediction models, such as ANN and tree-based models, offers significant potential for the OEM buyer’s purchasing strategy. For example, the ANN model can predict prices in real-time, allowing the purchasing strategy to be adapted quickly. In this study, various ML models such as tree-based and ANN have been applied for the prediction of steel material procurement prices. Tree-based models are commonly used in the literature as prediction tools due to their ease of creation, implementation, and interpretation. Although the prediction inference process of ANN models is defined as a black box, in the models developed for predicting steel material pro curement prices in this study, ANN has shown superior performance compared to tree-based methods in both testing and validation sets, hence an ANN-based model is recommended to the company. The Fig. 8. Fitted line graphics of ML algorithms for the validation set (a) for DT, (b) for GBT, (c) for RF, (d) for ANN with SCG, (e) for ANN with BR, (f) for ANN with LM. S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 11 experiments conducted with the tree-based methods clearly identified, the breakpoint values of the parameters affecting the prediction process, and the effect of parameter changes on the prediction value during the prediction process was easily examined. 5. Discussion In today’s changing world, purchasing functions need to be managed with agile and accurate decision-making. This study aims to formulate precise purchasing strategies and price predictions for an OEM auto motive company. In the previous researches, there are different price prediction studies with ML methods for different applications in the literature (Verlinden et al., 2008; Özcan and Fığlalı, 2014; Kuo and Xue, 1998; García and Kristjanpoller, 2019; Khoshalan et al., 2021). ANN is a widely used method for price prediction; however, the focal point of discussion pertains to the methods by which we can assess these pre dictions and effectively leverage them for purchasing objectives, given the paramount importance placed on strategies and decision-making processes within the realm of businesses. In this study, the ANN-based model demonstrated the most effective performance among the selected ML techniques for price prediction. To determine the optimal parameters of the ANN, the model was run with different number of hidden layers and number of neurons in hidden layers. The trained ANN and tree-based models were tested with different data set that were not used for training. However, because of the black box architecture of ANN, internal mechanisms are opaque and not easily understandable or explainable by researchers (Carabantes, Table 6 Business theories and ML models configurations. Decision Trees (DT) Gradient Boosted Trees (GBT) Random Forests (RF) ANN with SCG algorithm ANN with BR algorithm ANN with LM algorithm DT dynamically adjusts decision-making based on changing data patterns, reflecting organizational adaptability to remain responsive to shifting prices and estimations. GBT continuously improve predictive accuracy, reflecting the organization’s iterative learning processes. RF leverage diverse decision-making processes, akin to the organization’s exploration of various strategies. ANN dynamically adapt to changing data patterns, mirroring the organization’s continuous improvement efforts. BR balances model complexity and data fitting, reflecting the organization’s strategic resource allocation. LM lgorithm optimizes model parameters dynamically, akin to the organization’s real-time adaptation strategies. IPT illustrates DT’s systematic analysis of complex data relationships, fostering interpretability. It posits that decision trees utilize structured information processing to discern patterns and insights from diverse datasets. IPT enhances gradient boosting’s nuanced data analysis capabilities through ensemble learning. IPT highlights random forests’ efficiency in processing large data volumes and capturing relationships. IPT elucidates ANN’s ability to capture nonlinear relationships, enhancing prediction accuracy. IPT informs Bayesian regularization’s regularization techniques, preventing overfitting. IPT guides LM algorithm’s optimization, enhancing convergence speed and stability in training. DT facilitates transparent and interpretable managerial decision- making, aiding in initial exploratory analysis and feature identification. However, caution is advised in complex environments. GBT serves as a valuable tool for tasks necessitating high predictive accuracy, though its complexity demands expertise in model tuning and interpretation. RF excels in handling large, complex datasets with reduced overfitting compared to decision trees, finding applications in diverse domains such a big segmentation like automotive sector. SCG offer immense potential for managerial decision-making due to their ability to model complex relationships in data. They are widely used to uncover hidden patterns and trends. BR provides a principled approach to managerial decision-making by incorporating prior knowledge and balancing model complexity. It is particularly useful in scenarios with limited data or high feature dimensionality, such as financial predictions. LM commonly used in applications such as time series predictions, and improve decision accuracy and efficiency. However, careful initialization and monitoring are essential to ensure reliable results and prevent convergence issues. DT provide interpretable models that can effectively capture and represent complex pricing relationships. Their adaptability allows for real-time adjustments to changing market conditions, enhancing the model’s ability to accurately estimate prices. GBT offer superior predictive accuracy and robustness, making them suitable for precise price prediction tasks. Their iterative learning process allows for continual refinement of price prediction models, ensuring they stay up-to- date with changing market dynamics. RF offer versatility and scalability in price prediction tasks. Their ensemble nature allows them to capture complex relationships and interactions in the data, leading to accurate price predictions even in the presence of noise or outliers. ANN offer flexibility and adaptability in price prediction tasks. Their ability to capture nonlinear relationships allows them to model complex pricing dynamics effectively, providing accurate price estimates across a wide range of market conditions. BR offers stability and reliability in price prediction tasks. By effectively managing model complexity and preventing overfitting, BR ensures that the price prediction model remains robust and dependable, even in the presence of noisy or limited data. LM algorithm offers efficiency and effectiveness in price prediction tasks. Its optimization capabilities ensure fast convergence to an optimal solution, allowing for the development of accurate and reliable price prediction models in a timely manner. DT offer interpretability but may lack predictive power in complex environments. DT are prone to overfitting with noisy or unbalanced data, but their hierarchical structure allows for easy interpretation and understanding of decision-making processes. GBT mproves predictive accuracy through ensemble learning. It is less prone to overfitting compared to decision trees and tends to perform well in a wide range of applications. However, it requires careful tuning of hyperparameters to prevent model complexity and potential overfitting. RF mitigate overfitting and improve robustness. They offer high predictive performance and are less sensitive to noise and outliers compared to decision trees. However, they are computationally expensive and may suffer from reduced interpretability due to the ensemble nature of the model. ANN capture complex relationships in data. They excel at capturing nonlinear patterns and can handle large and diverse datasets effectively. However, they require careful tuning of hyperparameters and may be prone to overfitting, especially with small datasets. BR improves generalization and prevents overfitting. It effectively handles multicollinearity and reduces the risk of overfitting by incorporating prior knowledge about model parameters. However, it requires careful selection of regularization parameters and prior distributions. LM algorithm enhances convergence speed and stability. It efficiently optimizes model parameters by combining the advantages of gradient descent and Gauss- Newton methods. However, it may converge to local minima and requires a well-conditioned initial guess for optimal performance. S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 12 2020). This lack of transparency can hinder decision-makers under standing of the rationale behind the model’s predictions. For this reason, tree-based models have been integrated into the framework for price prediction. While it is acknowledged that their performance is not as optimal as that of ANN, the advantage lies in the ability to comprehend the decision criteria and weights that govern these models. It is notable that within these datasets, tree-based models elucidate the material features that exert a significant influence on price dynamics. The findings of this study establish the ANN model trained with the ANN-LM algorithm as the most effective approach for predicting steel material procurement prices, achieving an R-squared value of 0.953 on training and 0.836 on the validation set. However, the significance of this study extends beyond price prediction. In real-world applications, the integration of DCV and IPT demonstrates these predictions can be operationalized to drive strategic decisions. Using IPT, the organization can analyze large datasets and find practical insights by mimicking cognitive processes. Through this approach, price predictions derived from ML models are effectively contextualized, transforming raw pre dictions into strategic guidance. Simultaneously, DCV enables the or ganization to dynamically adapt purchasing strategies in response to market fluctuations, optimize supplier selection, and integrate sustain ability considerations based on ML and tree-based price prediction in sights. This dual theoretical lens ensures that price predictions are not only accurate but also actionable, fostering agility, sustainability, and competitive advantage in procurement practices. According to Table 6, the methods have been analyzed through the perspectives of the DCV and IPT. Subsequently, managerial impacts, integration of price esti mation models, and general comments have been incorporated. After the price predictions, an interdisciplinary perspective drawing from business theories becomes imperative for comprehensively sup porting these results, thereby facilitating informed decision-making for subsequent actions. This study employs DCV and IPT to improve the understanding and evaluation of price prediction model results. These theories provide frameworks for analyzing dynamic capabilities and information processing mechanisms within organizations. This enables a deeper insight into the implications of price predictions on strategic decision-making processes and aids in devising adaptive procurement strategies in response to changing market dynamics. This symbiotic relationship with price prediction models and business theories makes supply processes more predictable, cost savings, ultimately boosting competitiveness and efficiency. The results of this study make a substantial contribution to the literature by demonstrating the practical application of ML models, specifically ANN and tree-based algorithms, alongside business theories in the context of procurement price prediction. The findings emphasize the superior predictive accuracy of ANN while also highlighting the interpretability and transparency of tree-based models, offering valu able insights for decision-makers. These contributions are novel in that they address the dual challenge of improving prediction accuracy while maintaining transparency in decision-making processes. The integration of DCV and IPT with these ML models further enhances this research, providing a strategic framework for organizations to adapt to dynamic market conditions, optimize supplier selection, and make informed, agile purchasing decisions. By embedding these models within estab lished business theories, this study uniquely advances the application of ML in procurement, offering both theoretical and practical value. 6. Conclusions The integration of business theories with advanced price prediction models holds transformative potential for streamlining supply chain management and optimizing costs. Strategic product clustering, driven by ML algorithms, enhances supply chain management and cost opti mization. By deciphering patterns in consumption, pricing, and reve nues, organizations make judicious decisions in supplier selection and cost reduction. ML methodologies, including ANN and tree-based models, enhance price prediction accuracy. Carefully tuned ANN parameters and decision tree structures facilitate rapid predictions, empowering purchasing de partments to intervene swiftly when prices deviate. The incorporation of DCV and IPT enables organizations to adapt to changing market con ditions, facilitating real-time data analysis for more informed decision- making. This study underscores the pivotal role of ML in purchasing, enabling informed decision-making, optimal resource allocation, and strategic negotiations. This research paper makes a significant contribution to the field by integrating ML models with established business theories, such as the DCV and IPT. This novel integration not only improves the precision of purchasing price predictions through advanced ANN and tree-based algorithms but also offers strategic insights that align ML applications with procurement decision-making processes. By embedding these advanced models within the DCV and IPT frameworks, the study pro vides a more robust understanding of organizations can leverage dy namic capabilities and information processing mechanisms to make agile and informed procurement decisions, especially in the face of volatile market conditions. The study fills a crucial gap in the literature by presenting a cohesive model that merges theoretical perspectives with ML-based models, emphasizing how organizations can enhance purchasing strategies by harnessing both advanced computer science algorithms and business theories. This research also highlights the potential of these integrated approaches to improve supply chain resilience, decision-making accu racy, and responsiveness to market changes. This research lays the groundwork for future interdisciplinary in vestigations at the intersection of computer science and business do mains including different theories and different ML methods. Future research could delve deeper into the integration of computer science and business theories, extending beyond the DCV and IPT to include other relevant theories such as the Resource-Based View and Transaction Cost Economics. By combining these theories with advanced ML techniques, such as reinforcement learning, deep learning, and natural language processing, future studies could uncover novel approaches for enhancing procurement decision-making. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 13 Appendix Appendix A. . Graphical representation of different ANN structures with respect to MSE . Appendix B. . Graphical representation of different ANN structures with respect to MAE . Appendix C. . Graphical representation of different ANN structures with respect to r-squared . The ANN structure of the test number for Appendices A, B, and C can be followed by using the following pseudo-code: S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 14 % Define the sets for I and J I = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]; % Neurons in the first hidden layer J = [0, 2, 3, 4, 5, 6, 7, 8, 10, 12, 15]; % Neurons in the second hidden layer % Outer loop over J for j = J if j == 0 % Inner loop over I for i = I fprintf (’Test (%d, %d)\n’, i, j); end else % Inner loop over I for i = I fprintf (’Test (%d, %d)\n’, j, i); end end end Data availability The data that has been used is confidential. References Afrasiabi, S., Boostani, R., Masnadi-Shirazi, M. A., & Nezam, T. (2021). An EEG based hierarchical classification strategy to differentiate five intensities of pain. Expert Systems with Applications, 180, Article 115010. https://doi.org/10.1016/j. eswa.2021.115010 Akpinar, M., & Vincze, Z. (2016). The dynamics of coopetition: A stakeholder view of the German automotive industry. Industrial Marketing Management, 57, 53–63. https:// doi.org/10.1016/j.indmarman.2016.05.006 Allal-Chérif, O., Simón-Moya, V., & Ballester, A. C. C. (2021). Intelligent purchasing: How artificial intelligence can redefine the purchasing function. Journal of Business Research, 124, 69–76. https://doi.org/10.1016/j.jbusres.2020.11.050 Al-Qudah, R., Aloqaily, M., & Karray, F. (2022). Computer Vision-Based Architecture for IoMT Using Deep Learning. International Wireless Communications and Mobile Computing (IWCMC), 2022, 931–936. https://doi.org/10.1109/ IWCMC55113.2022.9825279 Arslankaya, S., & Toprak, Ş. (2021). Using Machine Learning and Deep Learning Algorithms for Stock Price Prediction. Uluslararası Muhendislik Arastirma ve Gelistirme Dergisi, 13(1), 178–192. https://doi.org/10.29137/umagd.771671 Aslam, H., Blome, C., Roscoe, S., & Azhar, T. M. (2018). Dynamic supply chain capabilities. Supply Chain, 38(12), 2266–2285. https://doi.org/DOI 10.1108/ IJOPM-09-2017-0555. Ay, M., Özbakır, L., Kulluk, S., Gülmez, B., Öztürk, G., & Özer, S. (2023). FC-Kmeans: Fixed-centered K-means algorithm. Expert Systems with Applications, 211, Article 118656. https://doi.org/10.1016/j.eswa.2022.118656 Bahrami, M., & Shokouhyar, S. (2022). The role of big data analytics capabilities in bolstering supply chain resilience and firm performance: A dynamic capability view. Information Technology & People, 35(5), 1621–1651. https://doi.org/10.1108/ITP- 01-2021-0048 Bansal, C., Deepa, P. R., Agarwal, V., & Chandra, R. (2024). A clustering and graph deep learning-based framework for COVID-19 drug repurposing. Expert Systems with Applications, 249, Article 123560. https://doi.org/10.1016/j.eswa.2024.123560 Bayram, S., & Çıtakoğlu, H. (2023). Modeling monthly reference evapotranspiration process in Turkey: Application of machine learning methods. Environmental Monitoring and Assessment, 195(1), 67. https://doi.org/10.1007/s10661-022-10662- z Bayram, S., Ocal, M. E., Laptali Oral, E., & Atis, C. D. (2016). Comparison of multi layer perceptron (MLP) and radial basis function (RBF) for construction cost estimation: The case of Turkey. Journal of Civil Engineering and Management, 22(4), 480. https:// doi.org/10.3846/13923730.2014.897988 Bentéjac, C., Csörgő, A., & Martínez-Muñoz, G. (2021). A comparative analysis of gradient boosting algorithms. Artificial Intelligence Review, 54, 1937–1967. https:// doi.org/10.1007/s10462-020-09896-5 Blumberg, J., & Thompson, G. (2022). Nonparametric segmentation methods: Applications of unsupervised machine learning and revealed preference. American Journal of Agricultural Economics, 104(3), 976–998. https://doi.org/10.1111/ ajae.12257 Bourdeau, M., Zhai, X. Q., Nefzaoui, E., Guo, X., & Chatellier, P. (2019). Modeling and forecasting building energy consumption: A review of data-driven techniques. Sustainable Cities and Society, 48, Article 101533. https://doi.org/10.1016/j. scs.2019.101533 Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32. https://doi.org/ 10.1023/A:1010933404324 Brida, J. G., Disegna, M., & Scuderi, R. (2014). Segmenting visitors of cultural events: The case of Christmas Market. Expert Systems with Applications, 41(10), 4542–4553. https://doi.org/10.1016/j.eswa.2014.01.019 Carabantes, M. (2020). Black-box artificial intelligence: An epistemological and critical analysis. AI & SOCIETY, 35(2), 309–317. https://doi.org/10.1007/s00146-019- 00888-w Čeperić, E., Žiković, S., & Čeperić, V. (2017). Short-term forecasting of natural gas prices using machine learning and feature selection algorithms. Energy, 140, 893–900. https://doi.org/10.1016/j.energy.2017.09.026 Chen, E., & He, X. J. (2019). Crude Oil Price Prediction with Decision Tree Based Regression Approach. Journal of International Technology and Information Management, 27(4), 2–16. https://doi.org/10.58729/1941-6679.1391 Chen, Y.-L., Hsu, C.-L., & Chou, S.-C. (2003). Constructing a multi-valued and multi- labeled decision tree. Expert Systems with Applications, 25(2), 199–209. https://doi. org/10.1016/S0957-4174(03)00047-2 Chirumalla, K. (2021). Building digitally-enabled process innovation in the process industries: A dynamic capabilities approach. Technovation, 105, Article 102256. https://doi.org/10.1016/j.technovation.2021.102256 Chowdhury, M. M. H., & Quaddus, M. (2017). Supply chain resilience: Conceptualization and scale development using dynamic capability theory. International Journal of Production Economics, 188, 185–204. https://doi.org/10.1016/j.ijpe.2017.03.020 Cui, C., & Fearn, T. (2017). Comparison of partial least squares regression, least squares support vector machines, and Gaussian process regression for a near infrared calibration. Journal of Near Infrared Spectroscopy, 25(1), 5–14. https://doi.org/ 10.1177/0967033516678515 Cui, M. (2020). Introduction to the K-Means Clustering Algorithm Based on the Elbow Method. Geoscience and Remote Sensing, 3, 9–16. https://doi.org/10.23977/ geors.2020.030102 Citakoglu, H. (2017). Comparison of artificial intelligence techniques for prediction of soil temperatures in Turkey. Theoretical and Applied Climatology, 130, 545–556. https://doi.org/10.1007/s00704-016-1914-7 Demir, V. (2022). Enhancing monthly lake levels forecasting using heuristic regression techniques with periodicity data component: Application of Lake Michigan. Theoretical and Applied Climatology, 148(3–4), 915–929. https://doi.org/10.1007/ s00704-022-03982-0 Demirbay, B., Kara, D. B., & Uğur, Ş. (2022). Multivariate regression (MVR) and different artificial neural network (ANN) models developed for optical transparency of conductive polymer nanocomposite films. Expert Systems with Applications, 207, Article 117937. https://doi.org/10.1016/j.eswa.2022.117937 Devi Priya, R., Sivaraj, R., Anitha, N., & Devisurya, V. (2022). Tri-staged feature selection in multi-class heterogeneous datasets using memetic algorithm and cuckoo search optimization. Expert Systems with Applications, 209, Article 118286. https://doi.org/ 10.1016/j.eswa.2022.118286 Fajar, M., & Nurfalah, Z. (2021). Hybrid Fourier Regression – Multilayer Perceptrons Neural Network for forecasting. Statistical Journal of the IAOS, 37(4), 1199–1204. https://doi.org/10.3233/SJI-210876 Foroozesh, N., Karimi, B., Mousavi, S. M., & Mojtahedi, M. (2023). A hybrid decision- making method using robust programming and interval-valued fuzzy sets for sustainable-resilient supply chain network design considering circular economy and technology levels. Journal of Industrial Information Integration, 33, Article 100440. https://doi.org/10.1016/j.jii.2023.100440 Galbraith, J. R. (1974). Organization Design: An Information Processing View. Interfaces, 4(3), 28–36. https://doi.org/10.1287/inte.4.3.28 García, D., & Kristjanpoller, W. (2019). An adaptive forecasting approach for copper price volatility through hybrid and non-hybrid models. Applied Soft Computing, 74, 466–478. https://doi.org/10.1016/j.asoc.2018.10.007 Gu, Q., Chang, Y., Xiong, N., & Chen, L. (2021). Forecasting Nickel futures price based on the empirical wavelet transform and gradient boosting decision trees. Applied Soft Computing, 109, Article 107472. https://doi.org/10.1016/j.asoc.2021.107472 S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 15 https://doi.org/10.1016/j.eswa.2021.115010 https://doi.org/10.1016/j.eswa.2021.115010 https://doi.org/10.1016/j.indmarman.2016.05.006 https://doi.org/10.1016/j.indmarman.2016.05.006 https://doi.org/10.1016/j.jbusres.2020.11.050 https://doi.org/10.1109/IWCMC55113.2022.9825279 https://doi.org/10.1109/IWCMC55113.2022.9825279 https://doi.org/10.29137/umagd.771671 https://doi.org/10.1016/j.eswa.2022.118656 https://doi.org/10.1108/ITP-01-2021-0048 https://doi.org/10.1108/ITP-01-2021-0048 https://doi.org/10.1016/j.eswa.2024.123560 https://doi.org/10.1007/s10661-022-10662-z https://doi.org/10.1007/s10661-022-10662-z https://doi.org/10.3846/13923730.2014.897988 https://doi.org/10.3846/13923730.2014.897988 https://doi.org/10.1007/s10462-020-09896-5 https://doi.org/10.1007/s10462-020-09896-5 https://doi.org/10.1111/ajae.12257 https://doi.org/10.1111/ajae.12257 https://doi.org/10.1016/j.scs.2019.101533 https://doi.org/10.1016/j.scs.2019.101533 https://doi.org/10.1023/A:1010933404324 https://doi.org/10.1023/A:1010933404324 https://doi.org/10.1016/j.eswa.2014.01.019 https://doi.org/10.1007/s00146-019-00888-w https://doi.org/10.1007/s00146-019-00888-w https://doi.org/10.1016/j.energy.2017.09.026 https://doi.org/10.58729/1941-6679.1391 https://doi.org/10.1016/S0957-4174(03)00047-2 https://doi.org/10.1016/S0957-4174(03)00047-2 https://doi.org/10.1016/j.technovation.2021.102256 https://doi.org/10.1016/j.ijpe.2017.03.020 https://doi.org/10.1177/0967033516678515 https://doi.org/10.1177/0967033516678515 https://doi.org/10.23977/geors.2020.030102 https://doi.org/10.23977/geors.2020.030102 https://doi.org/10.1007/s00704-016-1914-7 https://doi.org/10.1007/s00704-022-03982-0 https://doi.org/10.1007/s00704-022-03982-0 https://doi.org/10.1016/j.eswa.2022.117937 https://doi.org/10.1016/j.eswa.2022.118286 https://doi.org/10.1016/j.eswa.2022.118286 https://doi.org/10.3233/SJI-210876 https://doi.org/10.1016/j.jii.2023.100440 https://doi.org/10.1287/inte.4.3.28 https://doi.org/10.1016/j.asoc.2018.10.007 https://doi.org/10.1016/j.asoc.2021.107472 Hammann, D. (2023). Big data and machine learning in cost estimation: An automotive case study. International Journal of Production Economics, 109137. https://doi.org/ 10.1016/j.ijpe.2023.109137 Fiorentini, N., Pellegrini, D., & Losa, M. (2023). Overfitting Prevention in Accident Prediction Models: Bayesian Regularization of Artificial Neural Networks. Transportation Research Record, 2677(2), 1455–1470. https://doi.org/10.1177/ 03611981221111367 Hendry, L. C., Stevenson, M., MacBryde, J., Ball, P., Sayed, M., & Liu, L. (2019). Local food supply chain resilience to constitutional change: The Brexit effect. International Journal of Operations & Production Management, 39(3), 429–453. https://doi.org/ 10.1108/IJOPM-03-2018-0184 Hung, Y. H. (2022). Developing an improved ensemble learning approach for predictive maintenance in the textile manufacturing process. Sensors, 22(23), 9065. https://doi. org/10.3390/s22239065 Huang, Y., Huo, Z., & Fan, Y. (2024). DRA: A data reconstruction attack on vertical federated k-means clustering. Expert Systems with Applications, 250, Article 123807. https://doi.org/10.1016/j.eswa.2024.123807 Ikotun, A. M., Ezugwu, A. E., Abualigah, L., Abuhaija, B., & Heming, J. (2023). K-means clustering algorithms: A comprehensive review, variants analysis, and advances in the era of big data. Information Sciences, 622, 178–210. https://doi.org/10.1016/j. ins.2022.11.139 Ismkhan, H. (2018). I-k-means− +: An iterative clustering algorithm based on an enhanced version of the k-means. Pattern Recognition, 79, 402–413. https://doi.org/ 10.1016/j.patcog.2018.02.015 Jain, A. K., Mao, J., & Mohiuddin, K. M. (1996). Artificial neural networks: A tutorial. Computer, 29(3), 31–44. https://doi.org/10.1109/2.485891 Jaipuria, S., Jenamani, M., & Maria Arputham, R. (2015). The Strategic Procurement of Raw Material: A Case Study. International Journal of Procurement Management, 9. https://doi.org/10.1504/IJPM.2016.078689 Jones, A., & Knoppen, D. (2018). The role of strategic purchasing in dynamic capability development and deployment: A contingency perspective. International Journal of Operations & Production Management, 38(2), 446–473. https://doi.org/10.1108/ IJOPM-10-2015-0656 Justo-Silva, R., Ferreira, A., & Flintsch, G. (2021). Review on Machine Learning Techniques for Developing Pavement Performance Prediction Models. Sustainability, 13(9), 5248. https://doi.org/10.3390/su13095248 Kachamas, P., Akkaradamrongrat, S., Sinthupinyo, S., & Chandrachai, A. (2019). Application of Artificial Intelligent in the Prediction of Consumer Behavior from Facebook Posts Analysis. International Journal of Machine Learning and Computing, 9 (1), 91–97. https://doi.org/10.18178/ijmlc.2019.9.1.770 Kähkönen, A. K., Evangelista, P., Hallikas, J., Immonen, M., & Lintukangas, K. (2023). COVID-19 as a trigger for dynamic capability development and supply chain resilience improvement. International Journal of Production Research, 61(8), 2696–2715. https://doi.org/10.1080/00207543.2021.2009588 Kang, Z., Catal, C., & Tekinerdogan, B. (2020). Machine learning applications in production lines: A systematic literature review. Computers & Industrial Engineering, 149, Article 106773. https://doi.org/10.1016/j.cie.2020.106773 Kaparthi, S., & Bumblauskas, D. (2020). Designing predictive maintenance systems using decision tree-based machine learning techniques. The International Journal of Quality & Reliability Management, 37(4), 659–686. https://doi.org/10.1108/IJQRM-04- 2019-0131 Kayri, M. (2016). Predictive Abilities of Bayesian Regularization and Levenberg–Marquardt Algorithms in Artificial Neural Networks: A Comparative Empirical Study on Social Data. Mathematical and Computational Applications, 21, no. 2, 20. https://doi.org/10.3390/mca21020020 Khan, R. (2021). Artificial Intelligence and Machine Learning in Food Industries: A Study. Journal of Food Chemistry and Nanotechnology, 07(03). https://doi.org/ 10.17756/jfcn.2021-114 Khoshalan, H. A., Shakeri, J., Najmoddini, I., & Asadizadeh, M. (2021). Forecasting copper price by application of robust artificial intelligence techniques. Resources Policy, 73. https://doi.org/10.1016/j.resourpol.2021.102239 Kim, M., Im, I., & Han, S. (2016). Purchase Prediction by Analyzing Users’ Online Behaviors Using Machine Learning and Information Theory Approaches. Asia Pacific Journal of Information Systems, 26(1), 66–79. https://doi.org/10.14329/ apjis.2016.26.1.66 Kocabıyıkoğlu, A., Göğüş, C. I., & Gönül, M. S. (2016). Decision Making and the Price Setting Newsvendor: Experimental Evidence: Decision Making and the Price Setting Newsvendor. Decision Sciences, 47(1), 157–186. https://doi.org/10.1111/deci.12158 Kuo, R. J., & Xue, K. C. (1998). An intelligent sales forecasting system through integration of artificial neural network and fuzzy neural network. Computers in Industry, 37(1), 1–15. https://doi.org/10.1016/S0166-3615(98)00066-9 Kureljusic, M., & Karger, E. (2024). Forecasting in financial accounting with artificial intelligence – A systematic literature review and future research agenda. Journal of Applied Accounting Research, 25(1), 81–104. https://doi.org/10.1108/JAAR-06- 2022-0146 Li, A., Liu, M., & Sheather, S. (2023). Predicting stock splits using ensemble machine learning and SMOTE oversampling. Pacific-Basin Finance Journal, 78, Article 101948. https://doi.org/10.1016/j.pacfin.2023.101948 Li, H., Wu, Y., Cao, D., & Wang, Y. (2021). Organizational mindfulness towards digital transformation as a prerequisite of information processing capability to achieve market agility. Journal of Business Research, 122, 700–712. https://doi.org/10.1016/ j.jbusres.2019.10.036 Li, X., De Groot, M., & Bäck, T. (2021). Using forecasting to evaluate the impact of COVID-19 on passenger air transport demand. Decision Sciences, deci.12549. https:// doi.org/10.1111/deci.12549 Liu, C., Hu, Z., Li, Y., & Liu, S. (2017). Forecasting copper prices by decision tree learning. Resources Policy, 52, 427–434. https://doi.org/10.1016/j. resourpol.2017.05.007 Liu, J., Feng, Y., & Zhu, Q. (2021). Involving second-tier suppliers in Green supply chain management: Drivers and heterogenous understandings by firms along supply chains. International Journal of Production Research, 61, 4765–4785. https://doi.org/ 10.1080/00207543.2021.2002966 Liu, M. (2022). Assessing Human Information Processing in Lending Decisions: A Machine Learning Approach. Journal of Accounting Research, 60(2), 607–651. https://doi.org/10.1111/1475-679X.12427 Liu, R., Yang, B., Zio, E., & Chen, X. (2018). Artificial intelligence for fault diagnosis of rotating machinery: A review. Mechanical Systems and Signal Processing, 108, 33–47. https://doi.org/10.1016/j.ymssp.2018.02.016 Lourakis, M. I. A. (2005). A Brief Description of the Levenberg-Marquardt Algorithm Implemented by levmar.https://www.researchgate.net/publication/239328019_A_ Brief_Description_of_the_Levenberg-Marquardt_Algorithm_Implemened_by_levmar. Maghraoui, A. E., Hammouch, F.-E., Ledmaoui, Y., & Chebak, A. (2022). Smart Energy Management System: A Comparative Study of Energy Consumption Prediction Algorithms for a Hotel Building. 2022 4th Global Power, Energy and Communication Conference (GPECOM), 529–534. https://doi.org/10.1109/ GPECOM55404.2022.9815807. Martino, L., San Millán-Castillo, R., & Morgado, E. (2023). Spectral information criterion for automatic elbow detection. Expert Systems with Applications, 231, Article 120705. https://doi.org/10.1016/j.eswa.2023.120705 S.M. Masoumi N. Kazemi S.H. Abdul-Rashid Sustainable Supply Chain Management in the Automotive Industry: A Process-Oriented Review. Sustainability 2019 11. MacKay, D. J. C. (1992). Bayesian interpolation. Neural computation., 4(3), 415–447. Melville, N., & Ramirez, R. (2008). Information technology innovation diffusion: An information requirements paradigm. Information Systems Journal, 18(3), 247–273. https://doi.org/10.1111/j.1365-2575.2007.00260.x Møller, M. F. (1993). A scaled conjugate gradient algorithm for fast supervised learning. Neural Networks, 6(4), 525–553. https://doi.org/10.1016/S0893-6080(05)80056-5 Moskowitz, H., Drnevich, P., Ersoy, O., Altinkemer, K., & Chaturvedi, A. (2011). Using Real-Time Decision Tools to Improve Distributed Decision-Making Capabilities in High-Magnitude Crisis Situations*: Using Real-Time Decision Tools to Improve. Decision Sciences, 42(2), 477–493. https://doi.org/10.1111/j.1540- 5915.2011.00319.x Mukherjee, I., & Routroy, S. (2012). Comparing the performance of neural networks developed by using Levenberg–Marquardt and Quasi-Newton with the gradient descent algorithm for modelling a multiple response grinding process. Expert Systems with Applications, 39(3), 2397–2407. https://doi.org/10.1016/j.eswa.2011.08.087 Özcan, B., & Fığlalı, A. (2014). Artificial neural networks for the cost estimation of stamping dies. Neural Computing and Applications, 25(3–4), 717–726. https://doi. org/10.1007/s00521-014-1546-8 Öztemel, E. (2012). Yapay Sinir Ağları ((3rd ed.).). Papatya. Patel, P., Sivaiah, B., & Patel, R. (2022). Approaches for finding Optimal Number of Clusters using K-Means and Agglomerative Hierarchical Clustering Techniques. International Conference on Intelligent Controller and Computing for Smart Power (ICICCSP), 2022, 1–6. https://doi.org/10.1109/ICICCSP53532.2022.9862439 Paturi, U. M. R., & Cheruku, S. (2021). Application and performance of machine learning techniques in manufacturing sector from the past two decades: A review. Materials Today: Proceedings, 38(5), 2392–2401. https://doi.org/10.1016/j. matpr.2020.07.209 S.K. Paul R. Agarwal R.A. Sarker T. Rahman Supply Chain Risk and Disruption Management: Latest Tools 2023 Springer Nature Singapore Techniques and Management Approaches 10.1007/978-981-99-2629-9. Polikar, R. (2012). Ensemble learning. In C.a. Zhang, & Y. Ma (Eds.), Ensemble Machine Learning. Methods and Applications (1st ed.). New York: Springer. Krishna, R., Sathish, H., & Zhou, N. (2023). Forecasting Uncertainty Parameters of Virtual Power Plants Using Decision Tree Algorithm. Electric Power Components and Systems, 51(16), 1756–1769. https://doi.org/10.1080/15325008.2023.2205413 A. Ranganathan The Levenberg-Marquardt Algorithm 2004 https://sites.cs.ucsb.edu/ ~yfwang/courses/cs290i_mvg/pdf/LMA.pdf. Raviteja, B., Pandya, K. A., Khan, F., Tufail Khan, Z., Prajwal, R., & Kekatpure, A. (2022). Smart Supply Chain Management using Big Data Analysis and Machine Learning. International Conference on Edge Computing and Applications (ICECAA), 2022, 190–193. https://doi.org/10.1109/ICECAA55415.2022.9936359 Rehman, A. U., & Jajja, M. S. S. (2023). The interplay of integration, flexibility and coordination: A dynamic capability view to responding environmental uncertainty. International Journal of Operations & Production Management, 43(6), 916–946. https://doi.org/10.1108/IJOPM-11-2021-0708 Ruiz De Miras, J., Ibáñez-Molina, A. J., Soriano, M. F., & Iglesias-Parro, S. (2023). Schizophrenia classification using machine learning on resting state EEG signal. Biomedical Signal Processing and Control, 79, Article 104233. https://doi.org/ 10.1016/j.bspc.2022.104233 Sauer, P. C., & Seuring, S. (2019). Extending the reach of multi-tier sustainable supply chain management – Insights from mineral supply chains. International Journal of Production Economics, 217, 31–43. https://doi.org/10.1016/j.ijpe.2018.05.030 Saha, S., & Bandyopadhyay, S. (2012). Some connectivity based cluster validity indices. Applied Soft Computing, 12(5), 1555–1565. https://doi.org/10.1016/j. asoc.2011.12.013 Sakız, B., & Gencer, A. H. (2018). Forecasting the Bitcoin Price via Artificial Neural Networks. 438–444. https://doi.org/10.36880/C10.02070. Sharma, S., & Soni, H. K. (2020). Discernment of Potential Buyers Based on Purchasing Behaviour Via Machine Learning Techniques. IEEE International Conference on S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 16 https://doi.org/10.1016/j.ijpe.2023.109137 https://doi.org/10.1016/j.ijpe.2023.109137 https://doi.org/10.1177/03611981221111367 https://doi.org/10.1177/03611981221111367 https://doi.org/10.1108/IJOPM-03-2018-0184 https://doi.org/10.1108/IJOPM-03-2018-0184 https://doi.org/10.3390/s22239065 https://doi.org/10.3390/s22239065 https://doi.org/10.1016/j.eswa.2024.123807 https://doi.org/10.1016/j.ins.2022.11.139 https://doi.org/10.1016/j.ins.2022.11.139 https://doi.org/10.1016/j.patcog.2018.02.015 https://doi.org/10.1016/j.patcog.2018.02.015 https://doi.org/10.1109/2.485891 https://doi.org/10.1504/IJPM.2016.078689 https://doi.org/10.1108/IJOPM-10-2015-0656 https://doi.org/10.1108/IJOPM-10-2015-0656 https://doi.org/10.3390/su13095248 https://doi.org/10.18178/ijmlc.2019.9.1.770 https://doi.org/10.1080/00207543.2021.2009588 https://doi.org/10.1016/j.cie.2020.106773 https://doi.org/10.1108/IJQRM-04-2019-0131 https://doi.org/10.1108/IJQRM-04-2019-0131 https://doi.org/10.3390/mca21020020 https://doi.org/10.17756/jfcn.2021-114 https://doi.org/10.17756/jfcn.2021-114 https://doi.org/10.1016/j.resourpol.2021.102239 https://doi.org/10.14329/apjis.2016.26.1.66 https://doi.org/10.14329/apjis.2016.26.1.66 https://doi.org/10.1111/deci.12158 https://doi.org/10.1016/S0166-3615(98)00066-9 https://doi.org/10.1108/JAAR-06-2022-0146 https://doi.org/10.1108/JAAR-06-2022-0146 https://doi.org/10.1016/j.pacfin.2023.101948 https://doi.org/10.1016/j.jbusres.2019.10.036 https://doi.org/10.1016/j.jbusres.2019.10.036 https://doi.org/10.1111/deci.12549 https://doi.org/10.1111/deci.12549 https://doi.org/10.1016/j.resourpol.2017.05.007 https://doi.org/10.1016/j.resourpol.2017.05.007 https://doi.org/10.1080/00207543.2021.2002966 https://doi.org/10.1080/00207543.2021.2002966 https://doi.org/10.1111/1475-679X.12427 https://doi.org/10.1016/j.ymssp.2018.02.016 https://doi.org/10.1016/j.eswa.2023.120705 http://refhub.elsevier.com/S0957-4174(24)02853-7/h0335 https://doi.org/10.1111/j.1365-2575.2007.00260.x https://doi.org/10.1016/S0893-6080(05)80056-5 https://doi.org/10.1111/j.1540-5915.2011.00319.x https://doi.org/10.1111/j.1540-5915.2011.00319.x https://doi.org/10.1016/j.eswa.2011.08.087 https://doi.org/10.1007/s00521-014-1546-8 https://doi.org/10.1007/s00521-014-1546-8 http://refhub.elsevier.com/S0957-4174(24)02853-7/h0365 https://doi.org/10.1109/ICICCSP53532.2022.9862439 https://doi.org/10.1016/j.matpr.2020.07.209 https://doi.org/10.1016/j.matpr.2020.07.209 http://refhub.elsevier.com/S0957-4174(24)02853-7/h0385 http://refhub.elsevier.com/S0957-4174(24)02853-7/h0385 https://doi.org/10.1080/15325008.2023.2205413 https://doi.org/10.1109/ICECAA55415.2022.9936359 https://doi.org/10.1108/IJOPM-11-2021-0708 https://doi.org/10.1016/j.bspc.2022.104233 https://doi.org/10.1016/j.bspc.2022.104233 https://doi.org/10.1016/j.ijpe.2018.05.030 https://doi.org/10.1016/j.asoc.2011.12.013 https://doi.org/10.1016/j.asoc.2011.12.013 Advances and Developments in Electrical and Electronics Engineering (ICADEE), 2020, 1–5. https://doi.org/10.1109/ICADEE51157.2020.9368935 Shaoming, Z., Hui, X., & Linda, H. S. (2015). The Effects on Dynamic Capabilities on Value-Based Pricing and Export. Entrepreneurship in International Marketing, 25, 109–127. https://search-ebscohost-com.ezproxy.cc.lut.fi/login.aspx?direct=true &db=e000xww&AN=948293&site=ehost-live. Shobha, N., & Asha, T. (2017). Monitoring weather based meteorological data: Clustering approach for analysis. International Conference on Innovative Mechanisms for Industry Applications (ICIMIA), 2017, 75–81. https://doi.org/10.1109/ICIMIA.2017.7975575 Spurrier, J. D. (2003). On the null distribution of the Kruskal-Wallis statistic. Journal of Nonparametric Statistics, 15(6), 685–691. https://doi.org/10.1080/ 10485250310001634719 Susitha, E., Jayarathna, A., & Herath, H. M. R. P. (2024). Supply chain competitiveness through agility and digital technology: A bibliometric analysis. Supply Chain Analytics, 7, Article 100073. https://doi.org/10.1016/j.sca.2024.100073 Soubeiga, A., Antoine, V., Corteval, A., Kerckhove, N., Moreno, S., Falih, I., & Phalip, J. (2025). Clustering and Interpretation of time-series trajectories of chronic pain using evidential c-means. Expert Systems with Applications, 260, Article 125369. https:// doi.org/10.1016/j.eswa.2024.125369 Taylor, K. E. (2001). Summarizing multiple aspects of model performance in a single diagram. Journal of Geophysical Research: Atmospheres, 106(D7), 7183–7192. https:// doi.org/10.1029/2000JD900719 Truong, N.-S., Ngo, N.-T., & Pham, A.-D. (2021). Forecasting Time-Series Energy Data in Buildings Using an Additive Artificial Intelligence Model for Improving Energy Efficiency. Computational Intelligence and Neuroscience, 2021, 1–12. https://doi.org/ 10.1155/2021/6028573 Verlinden, B., Duflou, J. R., Collin, P., & Cattrysse, D. (2008). Cost estimation for sheet metal parts using multiple regression and artificial neural networks: A case study. International Journal of Production Economics, 111(2), 484–492. https://doi.org/ 10.1016/j.ijpe.2007.02.004 Yan, H., Jiang, Y., Zheng, J., Peng, C., & Li, Q. (2006). A multilayer perceptron-based medical decision support system for heart disease diagnosis. Expert Systems with Applications, 30(2), 272–281. https://doi.org/10.1016/j.eswa.2005.07.022 Zhang, J., Chan, F. T. S., & Xu, X. (2023). Data-driven analysis on optimal purchasing decisions in combined procurement. International Journal of Production Research, 61 (13), 4265–4278. https://doi.org/10.1080/00207543.2022.2051766 Zott, C. (2003). Dynamic capabilities and the emergence of intraindustry differential firm performance: Insights from a simulation study. Strategic Management Journal, 24(2), 97–125. https://doi.org/10.1002/smj.288 Zsidisin, G. A. (2003a). A grounded definition of supply risk. Journal of Purchasing and Supply Management, 9(5–6), 217–224. https://doi.org/10.1016/j. pursup.2003.07.002 Zsidisin, G. A. (2003b). Managerial Perceptions of Supply Risk. The Journal of Supply Chain Management, 39(1), 14–26. https://doi.org/10.1111/j.1745-493X.2003. tb00146.x Qin, Y., Xu, Z., Wang, X., & Skare, M. (2024). Artificial Intelligence and Economic Development: An Evolutionary Investigation and Systematic Review. Journal of the Knowledge Economy, 15(1), 1736–1770. https://doi.org/10.1007/s13132-023- 01183-2 Quinlan, J. R. (1986). Induction of decision trees. Machine Learning, 1, 81–106. https:// doi.org/10.1007/BF00116251 Wu, Y., Cheng-Liang, H., Quoc-Thang, P., & Yuan-Yao, L. (2022). Completed Review of Various Solar Power Forecasting Techniques Considering Different Viewpoints. Energies, 15(9), 3320. https://doi.org/10.3390/en15093320 Wu, X., Kumar, V., Ross Quinlan, J., Ghosh, J., Yang, Q., Motoda, H., McLachlan, G. J., Ng, A., Liu, B., Yu, P. S., Zhou, Z.-H., Steinbach, M., Hand, D. J., & Steinberg, D. (2008). Top 10 algorithms in data mining. Knowledge and Information Systems, 14(1), 1–37. https://doi.org/10.1007/s10115-007-0114-2 S. Mirasçı and A. Aksoy Expert Systems With Applications 266 (2025) 125986 17 https://doi.org/10.1109/ICADEE51157.2020.9368935 https://search-ebscohost-com.ezproxy.cc.lut.fi/login.aspx?direct=true%26db=e000xww%26AN=948293%26site=ehost-live https://search-ebscohost-com.ezproxy.cc.lut.fi/login.aspx?direct=true%26db=e000xww%26AN=948293%26site=ehost-live https://doi.org/10.1109/ICIMIA.2017.7975575 https://doi.org/10.1080/10485250310001634719 https://doi.org/10.1080/10485250310001634719 https://doi.org/10.1016/j.sca.2024.100073 https://doi.org/10.1016/j.eswa.2024.125369 https://doi.org/10.1016/j.eswa.2024.125369 https://doi.org/10.1029/2000JD900719 https://doi.org/10.1029/2000JD900719 https://doi.org/10.1155/2021/6028573 https://doi.org/10.1155/2021/6028573 https://doi.org/