Kampala City, Uganda’s capital, faces severe and escalating traffic congestion due to rapid urbanisation, population growth, and rising vehicle ownership, leading to substantial economic losses, increased emissions, and degraded quality of life. Traditional reactive traffic management systems are inadequate for addressing dynamic urban mobility patterns in resource constrained environments. This study develops an intelligent traffic congestion prediction framework using machine learning classification. A dataset of 500 observations from 15 major road segments is processed through rigorous preprocessing, domain-informed feature engineering (including capacity utilisation and flow efficiency), and ensemble classifiers to categorise congestion into four actionable severity levels: Low, Medium, High, and severe. Experimental evaluation on a stratified test set shows that XGBoost outperforms Random Forest, achieving 84.0% overall accuracy and the highest precision (0.808%), recall (0.840%), and F1-score (0.820%). Feature importance analysis highlights capacity utilisation, vehicle density, and flow efficiency as the dominant predictors, consistent with fundamental traffic flow theory. The proposed system establishes a scalable, interpretable foundation for proactive congestion management in developing cities. With future enhancements in real-world data integration and temporal-spatial modelling, it holds strong potential to support adaptive traffic control, incident response, and data-driven urban planning in Kampala and similar contexts.