Post by Stephan Fahrenkrog-Petersen

🧠 Ever wondered why some process discovery algorithms work great on one event log but fail on another? 🔍 In our new paper accepted for ICPM 2025, we introduce SHAining — a framework that applies Shapley value analysis to understand how different event log characteristics impact the performance of process mining algorithms. 🚀 We generated over 22,000 synthetic event logs and analyzed how features like trace length variance, activity entropy, or variant skewness influence metrics like fitness, precision, complexity, and execution time across algorithms like Inductive Miner, ILP Miner, and Split Miner. 📚 Check out the paper: https://lnkd.in/gT3HJdPa This work is a joint effort with Andrea Maldonado, Christian Frey, Anirudh A, Dr. Ludwig Zellner, and Thomas Seidl — and I’m super proud to be part of it!