A novel multi objective grey wolf optimization fuzzy miner for process discovery: Incorporating robustness and explainability in model evaluation.

Abstract

Process mining provides methodologies for analyzing, monitoring, and improving processes based on event logs. This study introduces Fuzzy Multi-Objective Grey Wolf Optimization (Fuzzy MOGWO), which integrates fuzzy modeling with a multi-criteria metaheuristic optimization approach. The proposed framework simultaneously optimizes six metrics: Fitness, Precision, Generalization, Simplicity, Robustness, and Explainability with the latter two newly proposed to evaluate noise resilience and analyst interpretability. A normalized scoring mechanism, based on the L₂ norm of all metrics, ensures balanced evaluation across objectives. Fuzzy MOGWO is benchmarked against Alpha Miner, Inductive Miner, and Fuzzy Miner using 10 synthetic noise-free logs, 10 synthetic noisy logs with 5-20% injected noise, and 3 real-world logs. Under noise-free conditions, it achieved a normalized score of 0.329, surpassing the best baseline (0.288) by 14.24%. In noisy environments, its score (0.440) exceeded the top competitor (0.378) by 16.40%. On real-world logs, it outperformed competitors in 4 out of 6 metrics, compared to 2 out of 6 for the PSO-based miner. These results demonstrate substantially improved effectiveness, robust performance in the presence of noise, and enhanced interpretability, establishing Fuzzy MOGWO as a comprehensive and reliable solution for challenging process discovery tasks.