All You Need Is Evolution: Rethinking Genetic Algorithms for Process Discovery

Kowalski, Frederik Hecter ; Nygård, Richard ; Axelsen, Jeppe Berg

4. semester

Data Science and Machine Learning, Msc.

2025

2025-06-12

41 pages

Process mining focuses on deriving accurate and understandable process models from event logs, with the goal of balancing key quality dimensions: fitness, precision, generality, and simplicity. Genetic algorithms have shown promise in navigating this trade-off due to their ability to explore a wide search space. However, traditional genetic process discovery approaches often suffer from long convergence times and typically fall short when compared to state-of-the-art algorithms such as the Inductive Miner and Split Miner. We revisit genetic process discovery and propose a novel enhancement to the evolutionary framework that significantly improves both its efficiency and effectiveness. Our approach introduces improvements such as the ability to generate good initial populations, highly efficient fitness estimation, and adaptive log filtering, which accelerates convergence and guides the search toward higherquality models. Through extensive evaluation on benchmark event logs, we demonstrate that our enhanced genetic algorithm achieves competitive performance, producing process models that rival those discovered by leading algorithms in terms of fitness, precision, and structural quality. These results suggest that, with the right improvements, genetic approaches can play a valuable role in the process discovery toolbox.

Download
View record in AAU Student Projects