Abstract:

To improve transportation safety management, 98 crew members from a railway transportation company, with 1 362 observations, were included in this study. A self-developed mental health questionnaire and facial emotion recognition technology based on deep learning were used, and independent-samples t tests, logistic regression, and random forest models were applied for analysis. The results show that mental health is specifically associated with operational performance, and facial emotional features can effectively predict psychological states. Significant differences are found in the emotional state dimension between the full-score and non-full-score groups, with the non-full-score group showing lower emotional state levels (p<0.05), whereas no significant differences are observed in the other dimensions. Education level and emotional state have significant positive effects on operational performance, whereas self-efficacy shows a negative effect. Age, years of service, mental fatigue, workload, and stress show no significant effects. In addition, attendance facial expression features show high predictive performance for all dimensions of mental health (R²>0.80), among which sadness, neutral expression, and surprise are the three most important emotional features. These findings indicate that facial emotional features are useful indicators for assessing the psychological states of heavy-haul locomotive crew members and provide empirical evidence for the development of dynamic safety monitoring and targeted psychological intervention.

Key words: heavy-haul locomotive crew, operational performance, mental health, facial emotion recognition, machine learning