Leveraging Anomaly Detection in Business Process with Data Stream Mining

Abstract

Identifying fraudulent or anomalous business procedures is today a key challenge for organisations of any dimension. Nonetheless, the continuous nature of business activities conveys to the continuous acquisition of data in support of business process monitoring. In light of this, we propose a method for online anomaly detection in business processes. From a stream of events, our approach extract cases descriptors and applies a density-based clustering technique to detect outliers. We applied our method to a real-life dataset, and we used streaming clustering measures for evaluating performances. Exploring different combinations of parameters, we obtained promising performance metrics, showing that our method is capable of finding anomalous process instances in a vast complexity of scenarios.