Graph-based feature extraction on object-centric event logs

Abstract

Abstract Process mining techniques have proven crucial in identifying performance and compliance issues. Traditional process mining, however, is primarily case-centric and does not fully capture the complexity of real-life information systems, leading to a growing interest in object-centric process mining. This paper presents a novel graph-based approach for feature extraction from object-centric event logs. In contrast to established methods for feature extraction from traditional event logs, object-centric logs present a greater challenge due to the interconnected nature of events related to multiple objects. This paper addresses this gap by proposing techniques and tools for feature extraction specifically designed for object-centric event logs. In this work, we focus on features pertaining to the lifecycle of the objects and their interaction. These features enable a more comprehensive understanding of the process and its inherent complexities. We demonstrate the applicability of our approach through its implementation in two significant areas: anomaly detection and throughput time prediction for objects in the process. Our results, based on four problems in a Procure-to-Pay process, affirm the potential of our proposed features in enhancing the scope of process mining. By effectively transforming object-centric event logs into numeric vectors, we pave the way for the application of a broader range of machine learning techniques, such as classification, prediction, clustering, and anomaly detection, thereby extending the capabilities of process mining.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 16 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.