ProM-Ex: An Explainable Framework for Anomaly Detection in Process Mining Using Large Language Models

Process mining involves analyzing business processes through event logs to discover, monitor, and optimize workflows, enhancing efficiency. In this domain, anomalies refer to deviations or unexpected patterns in process execution that may signal inefficiencies or errors. This work introduces a novel framework for explainable control flow anomaly detection in process mining, leveraging Explainable Artificial Intelligence (XAI) and Large Language Models (LLMs). The proposed framework Prom-Ex(Process Mining Explanation) aims to improve anomaly detection quality while providing interpretable results for end users. The evaluation utilizes real-world datasets, including the BPI Challenge 2013 and 2015 event logs. During the anomaly detection phase, n-grams are employed to identify patterns, which are subsequently analyzed using the Isolation Forest algorithm to detect anomalous behaviors. To explain the detected anomalies, the SHAP framework is applied, highlighting the contribution of individual features. These insights are then transformed into user-friendly textual explanations using the Gemini-Pro LLM, translating the root causes of anomalies into comprehensible messages. The framework's performance is assessed using the F1 score, measuring both anomaly detection accuracy and the clarity of explanations.