Narrative texts-based anomaly detection using accident report documents: The case of chemical process safety

This paper proposes a text‑mining workflow for learning from narrative accident reports. It extracts keywords from accident descriptions and applies a Local Outlier Factor (LOF) anomaly‑detection model to prioritize “unusual” accident conditions that may be underrepresented in frequency-based reviews. Using chemical process accident reports, it clusters anomalous cases and highlights distinctive risk keywords for each cluster. For Process Safety Management (PSM), this supports incident investigation and learning by surfacing rare-but-high-consequence patterns, improving the completeness of causal-factor discovery and the quality of corrective-action targeting.