SUPERVISED MODELS FOR ESTIMATING LINK-LEVEL TRAFFIC DENSITY USING TRAJECTORY DATA

Traffic congestion is a growing concern in rapidly expanding cities, particularly in contexts where conventional traffic monitoring systems provide limited spatial and temporal coverage. This challenge is especially visible in many cities of the Global South, where the scarcity of fine-grained data restricts detailed analysis of urban mobility at the road-segment level. This study examines the prediction of link-level traffic density in Abidjan using trajectory data collected from an app-based mobility platform during September 2023, comprising more than 11,000 observations aggregated over approximately 814 road segments, and supervised machine learning methods.Road segments are described through a combination of geometric, regulatory, and trajectory-based features, and several regression models are evaluated within a common experimental framework.The results indicate that reliable traffic density estimates can be obtained even in the absence of dense sensing infrastructure. Random Forest provides consistently accurate and stable predictions across heterogeneous traffic conditions. The analysis also suggests that regulatory characteristics, such as speed limits and road hierarchy, exert a stronger influence on traffic density than detailed geometric descriptors.These findings highlight the practical relevance of trajectory-based supervised learning as a flexible and affordable solution for traffic analysis and mobility planning in data constrained urban environments.