Explaining Decision-Making in Railway Route Decision Using SHAP And Machine Learning

This paper presents a novel approach to enhance the reliability and transparency of automated railway route decision-making by integrating Shapley Additive exPlanations (SHAP) with XGBoost. The framework provides both accurate predictions and interpretable explanations for railway route decisions, thereby addressing the critical need for explainable AI in safety-critical railway operations. Traditional machine learning models for railway route selection often function as "black boxes", limiting operator trust. To overcome this, we use SHAP to quantify the contribution of individual features (e.g., track changes, train status) to each decision. The research methodology involves collecting historical railway data, model training, SHAP integration, applying SHAP for feature attribution, and validating the system in real-world railway data. Preliminary testing in the UK Derby area shows improved explainability and decision reliability. The SHAP-ML framework not only predicts optimal routes but also provides detailed explanations that enable railway operators to understand the decision logic. The novelty of this research lies in applying SHAP to the railway domain for route decision-making, an area where its use is still underexplored. It shows promise for resilient and transparent railway automation.