Advancing Travel Time Prediction in Intelligent Transportation Systems Through Leaning-Based Uncertainty Quantification

Accurate travel time prediction is essential for road users, logistic operators, and public transportation planners for efficient route planning, fleet management, and timely arrival estimation of goods and passengers. This paper aims to advance travel time prediction by quantifying uncertainties arising from traffic demand, weather conditions, and incidents. The study uses a Bayesian Neural Network with Monte Carlo dropout to enhance the prediction accuracy, robustness, and reliability for applications in dynamic and complex traffic conditions. The results show that the dropout probability rate and activation functions are the two most important factors affecting the model’s performance in uncertainty quantification is particularly important for our downstream decision-making process. Additionally, uncertainty quantification leads to more explainable and actionable decisions. This approach can improve operational planning and ensure better service delivery and user satisfaction, highlighting its critical role in the advancement of intelligent transportation systems.