Discovering High-Quality Process Models Despite Data Scarcity

TL;DR

This paper demonstrates that object-centric process discovery algorithms can effectively learn high-quality process models with significantly less data, especially for complex concurrent processes, compared to traditional methods.

Contribution

It introduces an evaluation framework and classification of processes to assess data requirements, showing reduced data needs for object-centric discovery in complex scenarios.

Findings

Object-centric discovery reduces data needs for complex processes.

Traditional discovery struggles with concurrent subprocesses due to data scarcity.

Large-scale case study confirms effectiveness in manufacturing processes.

Abstract

Process discovery algorithms learn process models from executed activity sequences, describing concurrency, causality, and conflict. Concurrent activities require observing multiple permutations, increasing data requirements, especially for processes with concurrent subprocesses such as hierarchical, composite, or distributed processes. While process discovery algorithms traditionally use sequences of activities as input, recently introduced object-centric process discovery algorithms can use graphs of activities as input, encoding partial orders between activities. As such, they contain the concurrency information of many sequences in a single graph. In this paper, we address the research question of reducing process discovery data requirements when using object-centric event logs for process discovery. We classify different real-life processes according to the control-flow complexity within and between subprocesses and introduce an evaluation framework to assess process discovery algorithm quality of traditional and object-centric process discovery based on the sample size. We complement this with a large-scale production process case study. Our results show reduced data requirements, enabling the discovery of large, concurrent processes such as manufacturing with little data, previously infeasible with traditional process discovery. Our findings suggest that object-centric process mining could revolutionize process discovery in various sectors, including manufacturing and supply chains.