Learning from Change: Resilient Drift Detection and Adaptation in Process Mining

In dynamic environments where distributions of data vary with time, process mining models can be affected by concept drift. This paper proposes a mechanism of identifying and adjusting to the concept drift by using a statistical test, referred to as Adaptive Weighted Kolmogorov-Smirnov (KS) test. The proposed approach is experimented on a model known as the Adaptive Weighted Random Forest, and performance is tested in different drift conditions using artificial data produced by the STAGGER function. The metrics used to measure the performance of the method include accuracy, precision, recall, F1-score and model update latency. The findings show that the proposed technique is highly accurate in various settings and beats the traditional models in the detection and adaptation of concept drift. The technique will provide stability and low lag, which is appropriate in real-time implementation in extreme environments.