AI-DRIVEN INSIGHTS REPORT

Integrated Risk Analysis Framework

The proposed framework integrates HFACS for multi-level human factor structuring, SEM for linear transmission relationships, XGBoost for nonlinear risk modeling, and SHAP for explainable insights. This combines causal verification with explainable machine learning to systematically characterize risk evolution.

Human Factor Risk Co-occurrence Network

HFACS organizes human-related causes into four interrelated levels: Unsafe Acts, Preconditions, Unsafe Supervision, and Organizational Influences. The co-occurrence network reveals that risk propagates step-by-step from organizational influences to operational decision errors, forming a multi-level transmission pattern. Weakness in safety culture, poor supervision, and driver decision errors are frequently identified.

Figure 2: Co-occurrence network of HFACS human factor risk indicators. (Illustrative image link, replace with actual if provided)

Causal Path Diagram of HFACS Factors

Structural Equation Modeling (SEM) validates the HFACS framework, revealing linear transmission relationships. Environmental factors significantly influence 'preconditions for unsafe acts' but not directly 'unsafe acts,' suggesting an indirect impact by altering driver states. Operational management is a significant influence at the organizational level.

Figure 3: SEM causal path diagram. (Illustrative image link, replace with actual if provided)

XGBoost Superiority in Prediction

XGBoost demonstrates overall superior performance in predicting casualties and economic losses compared to Random Forest, Decision Tree, and Gradient Boosting. Its Mean Squared Error and Mean Absolute Error are significantly lower, indicating higher prediction accuracy and strong generalization ability.

Impact of Severe Weather on Driver Decisions

Global SHAP analysis reveals that severe weather strongly interacts with driver decision errors. While driver violations and operation errors contribute more under normal weather, extreme severe weather amplifies the SHAP values for driver decision errors, making them the most critical factor in accidents under harsh conditions.

Figure 4: Global SHAP analysis. (Illustrative image link, replace with actual if provided)

Criticality of Driver Decision Errors in Local Accidents

Local SHAP analysis for single accident cases confirms that driver decision errors play a dominating role, especially under severe weather, exhibiting overwhelmingly higher SHAP values. This implies that severe weather profoundly reconstructs the human-factor pattern, elevating 'decision errors' to a core trigger.

Figure 5: Local SHAP analysis. (Illustrative image link, replace with actual if provided)

Nonlinear Amplification of Risk under Severe Weather

Severe weather doesn't just add risk linearly; it significantly reshapes driver perception, cognitive load, and decision stability. Environmental factors act as external triggers, vehicle factors as physical amplification channels, human factors as the direct core of failure, and management factors as long-term latent sources of risk amplification. This complex interplay can lead to system instability and severe consequences.

Proactive Risk Prevention & Training

This research highlights that severe weather significantly amplifies the risk of driver decision errors, making them the core trigger of accidents in adverse conditions. In contrast, under normal weather, operational deficiencies are more dominant. This understanding is crucial for traffic safety management and accident forecasting.

Future work will integrate more detailed environmental factors and driver psychological states, alongside real-time weather and driving behavior monitoring, to enhance predictive accuracy. Enhanced policy making and driver training are vital to reduce hazardous chemical tanker accidents, particularly focusing on preparing drivers for dynamic and intricate weather conditions.