Effects of Exogenous Factors and Bayesian-Bandit Hyperparameter Optimization in Traffic Forecast Analysis

Traffic congestion is a significant urban challenge with far reaching effects on society, the economy, and the environment. Accurate short-term traffic flow prediction is essential for the development of Intelligent Transport Systems (ITSs) to manage traffic flows efficiently. However, predicting traffic flow is complex due to spatiotemporal dependencies and external factors like weather, peak hours, and holidays. While several prediction methods capture spatiotemporal dependencies, they overlook external factors. This study emphasizes the significance of understanding exogenous influences on traffic flow prediction. The proposed system utilizes the additive regressive time series forecasting model—Prophet, which accommodates selected exogenous factors, to best capture the stochastic nature of traffic. In order to increase precision, it also integrates Bayesian-directed search techniques for optimizing the most crucial model hyperparameters. The empirical findings demonstrate that including external factors and their effects in the model improves forecast accuracy by capturing the traffic trend over weekends and seasonal holidays.