EURO 2024 Copenhagen
Abstract Submission

EURO-Online login

3490. Balancing Realism and Simplicity: A Synthetic Dataset Approach for Comprehensive Model Benchmarking in Prescriptive Process Monitoring

Invited abstract in session TB-27: Advances in Process Mining, stream Mathematical Optimization for XAI.

Tuesday, 10:30-12:00
Room: 047 (building: 208)

Authors (first author is the speaker)

1. Jakob De Moor
FEB, KU Leuven

Abstract

Prescriptive Process Monitoring (PresPM) is an emerging field dedicated to devising methods for suggesting interventions in business processes to optimize them. Machine learning models are applied using event logs to attain desired outcomes for each case of a process. Current research in PresPM reveals two significant gaps. The first involves the underutilization of synthetic data in existing literature. Synthetic datasets offer easy customization and the ability to generate ground truth counterfactuals for a given case, providing precise model evaluation and explainability of an intervention recommendation. However, these datasets are often overlooked and can be unrealistic. The second gap revolves around the often exclusive focus on a narrow range of machine learning approaches and intervention types, restricting a holistic understanding of PresPM methodologies. To bridge these gaps, we conduct a thorough comparison of machine learning methods using a synthetic dataset mimicking a local bank's loan application process. This dataset strikes a balance between real-life complexity and simplicity, allowing meaningful comparisons. Our approach involves variants of the established methods Causal Inference and (online) Reinforcement Learning and we compare insights obtained from the synthetic dataset with historical datasets to assess consistency when applied to real-world data. This research aims to contribute to a more comprehensive understanding of current methodologies.

Keywords

Status: accepted

Back to the list of papers