Leveraging Machine Learning for Demand Prediction in On-Demand Public Shuttle Services: A Case Study

This study explores the use of machine learning for demand prediction in on-demand public shuttle services. Using historical ridership data enriched with contextual factors such as time of day, day of week, weekend indicators, weather, and special events, the research evaluates multiple ML models including Linear Regression, Random Forest, and XGBoost. The dataset was chronologically split (80–20) to preserve temporal patterns, ensuring realistic forecasting scenarios. Among the models tested, the Random Forest Regressor demonstrated the best performance, achieving a Mean Absolute Error (MAE) of 5.25 and a Root Mean Squared Error (RMSE) of 11.65. Results show that temporal features (hour, day, weekend) are the strongest predictors of shuttle demand, while contextual and spatial features add further predictive value. The findings highlight the potential of lightweight, interpretable ML approaches to improve fleet allocation, reduce passenger wait times, and optimize operational costs in smart urban mobility systems.