Optimized supervised machine learning for accurate prediction of periods in Türkiye's heritage minarets

Abstract

Historical masonry minarets, known for their tall and slender forms, are especially susceptible to environmental and seismic impacts because of their distinct structural characteristics. Traditional methods, such as three-dimensional numerical modeling, are widely used to evaluate the stability of these structures. However, the complex, heterogeneous materials common in historical construction often lead to significant differences between simulated predictions and actual dynamic behaviors, posing challenges to accurately assessing their stability. This study addresses these issues by introducing a supervised machine learning (SML) approach specifically designed to predict the fundamental period of 27 historical minarets in T & uuml;rkiye. Unlike conventional techniques that depend on extensive field testing or highly detailed numerical models, this SML model utilizes straightforward geometric (such as equivalent height and diameters) and material parameters (Young's modulus and mass density) to achieve accurate predictions with high reliability. Additionally, the input vectors are expanded to include slenderness parameters, significantly enhancing prediction accuracy. Three optimized SML functions are systematically evaluated, with the Grid Search method identified as the most effective approach for this application. The inclusion of slenderness parameters and the use of the Grid Search method yields exceptional prediction performance, demonstrating the outstanding of the proposed methodology compared to some existing empirical equations when achieving prediction error margins below 20%. This framework offers a practical, non-invasive tool for analyzing the dynamic stability and resilience of culturally significant structures, providing a modern, efficient solution for heritage conservation.