Süreç Madenciliğinde Tahmine Dayalı Süreç İzleme Tekniklerinin Kalitesinin Açıklanabilir Yapay Zeka Yöntemleri İle Değerlendirilmesi

Abstract

Predictive process monitoring (PPM) techniques used in process mining aim to predict future attributes of a process, such as process outcome, next activity to be executed, or remaining work time. Many existing PPM approaches utilize machine learning (ML) based prediction models. The increasing efficiency and accuracy of ML models is often combined with increasing complexity, compromising the understandability of these models. Explainable artificial intelligence (XAI) methods have emerged to provide users with explanations of the reasoning process of an ML model. However, the choices made and the techniques used within the scope of PPM also affect the resulting explanations. This study synthesizes the literature to construct a foundational understanding of explainable artificial intelligence in the field of predictive process monitoring. The types of predictive models used, XAI techniques applied, and the empirical evaluation and validation of XAI techniques are analyzed. The study highlights the evolving aspects of XAI in PPM, demonstrating the importance of explainability to end-users, the potential of textual data to enrich predictive models, and advancements in both conceptual and empirical frameworks. The insights obtained can inform the design of more interpretable and trustworthy artificial intelligence systems and facilitate the adoption of these systems in critical business process management applications. In this study, XAI supported predictive process monitoring steps were derived in the light of the information obtained after the literature was synthesized. Within the scope of the case study, by applying the determined predictive process monitoring techniques to a selected event log using these steps, the quality of the predictions was measured with performance metrics and explanations were made with the determined explainable artificial intelligence methods. Then, together with the results of the performance metrics and explanations, the models were evaluated and necessary improvements were made with hyperparameter optimization and the obtained results were shared. This study provides comprehensive information on the preparation, training, evaluation of predictive process monitoring models and the creation of explanations with XAI methods.