Intelligent Transport Systems for Urban Bus Networks Using Multilayer Perceptron and XGBoost Regression for Real-Time Operational Capability

This paper provides a comprehensive study on how to improve an urban bus transportation system using machine learning techniques to increase operational efficiency and service quality. The research examines the relationships between important operational parameters, including route length, average speed, headway intervals, and fleet size, through empirical data analysis of 28 bus routes. To address the complex interdependencies in urban public transportation systems and predict headway optimization, two advanced machine learning models, Multi-Layer Perceptron (MLP) and XG Boost Regression, were implemented. The dataset revealed significant variations in operational parameters, with route length ranging from 6.79 km to 34.42 km, average speed ranging from 14.12-34.67 km/h, and headway ranging from 3.27-18.05 minutes. Correlation analysis identified a strong negative relationship (-0.81) between headway and fleet size, indicating that more frequent services require larger fleet allocations. On the training data, the R2 of the MLP model was 0.857, however on the test data, it did not perform well (R² = 0.603), indicating potential overfitting issues. In contrast, the XG Boost regression model performed well with perfect training accuracy (R² = 1.000) and maintained strong generalization ability (R² = 0.901) on the test data. These findings highlight the effectiveness of gradient boosting techniques in capturing complex nonlinear relationships within transportation systems. This study adds to the expanding body of knowledge in intelligent transportation systems by providing practical insights to transportation operators to optimize resource allocation, improve service reliability, and increase passenger satisfaction while maintaining economic sustainability in urban public transportation operations.