Seeing the Forest and the Trees: Using Machine Learning to Categorize and Analyze Incident Reports for Alberta Oil Sands Operators

This foundational study analyzes 15,000 incident reports from five Alberta oil sands companies using supervised machine learning to create a standardized, cross-company risk matrix. The Linear SVC classifier achieved 89.98% accuracy for risk score assignment and 85.68% for PSM element classification. The research demonstrates how ML can harmonize disparate incident reporting practices across organizations, enabling industry-wide trend analysis. This is directly relevant to Incident Investigation in PSM because it automates the classification and severity scoring of incident reports, enabling safety teams to identify recurring patterns and prioritize investigations more effectively across the oil sands sector.