Graph-Based Feature Extraction on Object-Centric Event Logs

Abstract

Process mining techniques are widely used to uncover performance and compliance problems. However, the traditional focus on a single object type (i.e., case) is a limiting factor when considering real-life information systems. Therefore, there is an increased interest in object-centric process mining. This paper proposes a graph-based approach for feature extraction on object-centric event logs. The conversion of the event log to a set of numeric vectors is the starting point for the application of any machine learning technique (classification, prediction, clustering, anomaly detection) on top of the event data. However, while feature extraction on traditional event logs is established, object-centric event logs are more difficult to encode in numeric features because the events are related to several interconnected objects. Here, we try to close this gap and propose techniques and tools implementing feature extraction on object-centric event logs. The usefulness of the proposed features is discussed on top of four problems in a Procure-to-Pay process.