An efficient drift detection approach using data entropy in business processes

Business processes evolve due to adaption to variable environmental conditions. It is vital for organization managers to discover process changes in order to prevent performance issues in organizational processes. These changes may occur suddenly, gradually, periodically, or incrementally. This paper proposes an algorithm for detecting concept drift in event logs for sudden drifts, based on the entropy of trace variants in the execution of processes. In the proposed approach, two sliding window approach is leveraged to extract features of trace variants is given intervals. By counting the number and variety of traces in the reference and detection windows, two feature vectors are created for each window. Comparing two feature vectors that are obtained from two sliding windows by the entropy of feature vectors and SymGtest distance function of trace variants, possible drifts are detected. Experiments on synthetic databases show the accuracy of the method and its superiority to state-of-the-art methods.