Hybrid optimization-predictive analytics model for dynamic rake scheduling and ETA prediction in rail freight operations

Rail freight is vital for economic growth, offering cost-effective transportation solutions. However, their privatisation within incentive regulation or public-private partnerships often reduces efficiency due to ineffective fleet planning, and government control. This paper presents a hybrid approach to address fleet planning challenges in rake (freight train) scheduling and rescheduling dynamically for private Rail Freight Operators, highlighting the critical role of optimisation model and real-time estimated time of arrival (ETA) prediction model. The first phase involves developing an integer linear programming (ILP) model for optimal rake assignment and scheduling for maximum profit generation. In the second phase, a data-driven machine learning-based regression predictive model for ETA predictions is developed using GPS data. The Upgraded predictive model outperformed the base model, with XGBoost achieving 6.15 h Root Mean Square Error for Loaded direction and Feature-weighted K-NN 9.09 h for Empty direction. Furthermore, this study integrates ETA prediction with the ILP model for real-time rake rescheduling, enhancing profit, operations, and customer satisfaction through accurate delivery estimates. We validated using real-time data from an Indian private Rail Freight Operator within the Indian Railways network. The ILP model increased RFO monthly trips by 20%, enhancing profit, with ETA predictions over 95% accurate.