Predicting Service Efficiency and Revenue Patterns in UK National Rail Transport Using Machine Learning Techniques

Abstract

This study applies machine learning techniques to predict service efficiency and revenue patterns within the UK National Rail transport system. Using a curated rail transport dataset, the study investigates operational and financial factors associated with journey outcomes and ticket pricing. Data preprocessing included feature engineering, outlier treatment using the interquartile range method, log transformation, encoding, class balancing using SMOTE, and Mutual Information-based feature selection. For service efficiency prediction, XGBoost, CatBoost, LightGBM, and Support Vector Machine models were evaluated. LightGBM achieved the highest classification accuracy of 93%, with strong precision, recall, and F1-scores for the major journey-status classes, although performance was weaker for the minority Cancelled class. For revenue prediction, Polynomial Regression outperformed Ridge and Lasso Regression, achieving an MAE of 5.60, RMSE of 11.78, and R² of approximately 0.86. The findings demonstrate the potential of supervised machine learning and nonlinear regression techniques to support operational planning, service reliability, and revenue forecasting within the UK rail sector.