On the use of online clustering for anomaly detection in trace streams

Identifying anomalies in business processes is a challenge organizations face daily and are critical for their operations' data flow, whether public or private. Most current techniques face this challenge by requiring prior knowledge about business process models or specialists intervention to support the usage of state of the art methods, such as supervised machine learning. Also, the techniques tend to perform offline towards achieving consistent predictive results. In this work, we propose identifying the effectiveness of an online clustering method, particularly Autocloud. This algorithm is able to perform anomaly detection in trace streams meeting real-life requirements. Autocloud is an autonomous, evolutionary, recursive online clustering algorithm that requires little memory to provide insights from anomalous patterns in real-time. Moreover, this clustering algorithm does not require previous training or even prior knowledge from the application domain. Experiments were carried out with six process models, six different anomalies over 1,000, 5,000, and 10,000 event traces, generating a total of 630 datasets. The experiments confirmed the algorithm's ability to detect anomalies in those event traces, paving the way for more reliable information systems grounded on an automatic conformance checking of desirable business process execution.