The railway network constitutes a vital component of public transportation in many countries, serving millions of passengers and transporting significant volumes of freight. Nevertheless, a persistent challenge within this system is the frequent occurrence of train delays, which arise from diverse causes and result in financial losses, passenger dissatisfaction, and diminished trust among users. Consequently, enhancing operational efficiency and minimising delays has become a central objective for transportation planners and policymakers. In addressing this issue, the present study applies machine learning algorithms (MLAs), specifically multilayer perceptron (MLP) neural networks and the adaptive neuro-fuzzy inference system (ANFIS), to predict potential defects in railway vehicles and improve maintenance and repair strategies within the Iranian railway network. The findings reveal that ANFIS achieves superior predictive accuracy. Building on this, a mathematical model in combination with the Particle Swarm Optimization (PSO) algorithm was developed to optimise train allocation across stations and generate schedules aimed at reducing delays. The employed algorithms proved to be highly effective for predictive maintenance and repair of railway vehicles, ultimately contributing to delay reduction within the railway system.