Process Mining-Based Approach for Incremental and Recurring Concept Drift Detection

Concept drift presents a major challenge in dynamic operational environments, where evolving process behavior can lead to degradation in the performance of predictive and analytical systems. In the context of business processes, such drift often manifests as changes in the control flow, which can be either incremental—gradually evolving—or recurring—reappearing over time. This paper proposes a novel framework for detecting and localising both incremental and recurring concept drifts in the control-flow perspective using event logs generated from operational processes. The proposed technique employs a sliding window-based segmentation of logs, followed by the generation of control-flow models using inductive process mining algorithms. A hybrid drift detection strategy integrates conformance checking, graph similarity measures, and adaptive replay buffers to identify and characterize drift patterns over time. Experimental results on benchmark and synthetic datasets demonstrate the effectiveness of the proposed approach in terms of detection accuracy, localization precision, and resilience to noise. This research contributes toward robust, interpretable, and real-time process monitoring systems that adapt to evolving operational dynamics.