Simultaneous Process Drift Detection and Characterization with Pattern-Based Change Detectors

Traditional process mining approaches learn process models assuming that processes are in steady-state. This does not comply with the flexibility and adaptation often requested for information systems and business models. In fact, these approaches should discover variations to adapt to new circumstances, which is a peculiarity that conventional change analysis based on time-series, could not provide, because the processes are complex artifacts. This problem can be handled with change-aware structured representations, such as those typically used for network data. In this paper, we propose a novel pattern-based change detection (PBCD) algorithm for discovering and characterizing changes in event logs encoded as dynamic networks. In particular, PBCDs are unsupervised change detection methods, based on observed changes in sets of patterns observed over time, which are able to simultaneously detect and characterize changes in evolving data. Experimental results, on both real and synthetic data, show the usefulness and the increased accuracy with respect to state-of-the-art solutions.