From Anger Detection to Intensity Modeling: A Two-Stage Machine Learning Approach Using Driving Performance and Physiological Signals.

OCCUPATIONAL APPLICATIONSThis study demonstrated that driver anger can be feasibly modeled across three intensity levels using combined driving performance metrics and physiological signals. A two-stage machine learning framework, which first determines anger presence and then classifies its intensity, substantially improved accuracy and reduced neutral state misclassification compared to a single-stage four-class model. These findings have direct implications for reducing safety risks and promoting long-term health for workers whose jobs involve extensive driving (e.g., commercial drivers, bus operators), who encounter anger-eliciting situations more frequently than non-occupational drivers. Integrating anger-intensity detection into driver monitoring systems can enable adaptive, context-aware assistance systems that consider both intervention timing and emotional intensity. Aggregated emotion-intensity information may also inform operational decisions (e.g., dispatch assignments and break scheduling) by identifying periods when drivers may benefit from reduced demands or modified tasks. These implications can enhance fleet safety programs and support worker well-being by reducing exposure to emotionally demanding conditions.