Analyzing results of business process automation with machine learning methods

Abstract

Production and management systems and enterprise resource management systems are constantly creating data. Businesses use integrated systems to monitor all processes such as sales, planning, production and logistics systems. During the follow-up and use of these systems, office employees perform routine, repetitive and non-value-added transactions. Manual processes such as invoice entries, sales data entry, and order transfer significantly reduce employee satisfaction and cause some personal errors. Due to the developing requirements, the concept of Robotic Process Automation (RPA), which can operate like humans in many programs and customer systems, have emerged in recent years. There are softwares that work as a white collar employee in enterprises to perform RPA-defined, non-interpretation-based, rule-based and standard tasks. In this study, we study the task of uploading invoices to the customer system, that is one of the standard and routine transactions in Logistics Processes. These tasks are automated with RPA. Software robots repeat the processes and purify the process from non-value-added transactions. However, software robots receive some errors in the processes. Data mining methods are used in this study, in order to examine the software outputs and RPA errors. Reports on the results of RPA were analyzed with various machine learning methods using the WEKA software. As a result of the study, the J48 algorithm with an F-value of 75% gave the best result in the estimation of RPA outputs. In future studies, analyses will be made to examine and eliminate the root causes of the errors.