Trust-constrained multi-objective hybrid ensemble framework for spatiotemporal passenger demand prediction for autonomous taxi systems

Accurate taxi-demand forecasting has become a core requirement for modern smart-city planning and Intelligent Transportation Systems (ITS). As autonomous vehicles, especially Autonomous Taxis (ATs), move toward mainstream deployment, cities need reliable predictions to manage wait times, balance supply with demand, and support efficient fleet operations. This study introduces a trust-constrained hybrid ensemble designed to strengthen AT pick-up demand prediction. The framework merges the distinct advantages of TabNet, a Multi-Layer Perceptron, and Stochastic Gradient Boosting to capture complex spatiotemporal patterns. Instead of assigning equal or heuristic weights, the model applies a multi-objective optimization routine that jointly considers 𝑅$^2$, MAE, and Huber loss within a trust-region solver to improve stability and generalization. The evaluation uses a large-scale dataset of more than 4,700 taxis in Bangkok, enriched with spatiotemporal attributes and synchronized weather variables such as temperature, humidity, rainfall, wind speed, and atmospheric pressure. Simulation results demonstrate that the proposed ensemble consistently outperforms standalone models and equal-weight baselines, achieving a testing RMSE of 9.69, MAE of 5.97, and 𝑅$^2$ of 0.924. ... mehrThese findings highlight the framework’s predictive capability and practical potential for improving fleet management, optimizing vehicle deployment, and reducing idle times, thereby supporting data-driven decision-making for efficient and intelligent AT systems.