Forecasting Urban Traffic Congestion using Bayesian-Driven Models

One of the challenges of urbanizing African cities in terms of mobility is traffic congestion.Congestion, long queues, and unpredictable travel times in Calabar, Nigeria, are a daily experience, especially in commercial areas and along major roads.In this study, congestion states are predicted using Gaussian Process Regression (GPR) and a Relevance Vector Machine (RVM), based on field data from 20 large corridors.The findings indicate considerable variation in congestion rates across places.Post hoc Tukey HSD tests indicate that congestion is intermittent and is instigated by localized pinpoints of congestion at the Harbor, Tinapa, Main Avenue, Goldie, and Ettagbor crossings.Ekpo Abasi, Uwanse, and Goldie, on the other hand, were doing fairly well regarding delay and queue length.Vehicle arrivals were the most important feature, with the strongest linear correlations with total delay (r = 0.669) and queue length (r = 0.858), and traffic demand showed a positive linear association with the volume-to-capacity ratio (r = 0.568).GPR was more effective than RVM in all the congestion indicators.GPR obtained R² values of 0.876 (total delay), 0.966 (queue length), and 0.735 (VCR) on the test set with well-calibrated 95% credible interval coverage (94.1%, 93.0%, 90.2%).The locations of congestion were detected as clusters in the spatial maps around markets, schools, and transport termini.These findings support congestion hotspot-specific interventions and endorse probabilistic modeling, particularly GPR, for traffic prediction in data-sparse African cities.This study provides evidence-based recommendations for transportation policy, route planning, and signal optimization in Calabar.