Congestion-Based Traffic Control with Dynamic Algorithms and Supervised Machine Learning

Urban traffic management has become increasingly complex due to growing vehicle density and dynamic road conditions. This project introduces a hybrid machine learning framework for intelligent traffic light control, integrating dynamic algorithms like adaptive XGBoost (shallow and deep), HistGradientBoosting, MLP, and LightGBM for congestion classification, alongside optimization techniques. The system analyzes structured traffic parameters, including timestamp-derived features (day, hour, minute), IR presence across four lanes, vehicle count, average speed, detected vehicle types, vehicle density percentage, and congestion level. Supervised machine learning models are employed to classify congestion levels, while the optimization approach dynamically adjusts signal timing strategies. This hybrid model offers predictive accuracy and adaptive signal control to enhance traffic flow. The architecture is implemented using a Flask-based backend and a web interface with HTML, CSS, and JavaScript, enabling users to register, log in, and perform traffic prediction or classification. The system improves signal coordination, reduces congestion, and enhances decision-making through intelligent automation. Experimental evaluation demonstrates improved classification performance and optimized traffic signal control strategies compared to conventional static timing systems.

Keywords: Traffic Light Control, Hybrid Machine Learning, Adaptive XGBoost, HistGradientBoosting, MLP, LightGBM, Congestion Classification, Traffic Optimization, Signal Control, Reinforcement Learning, Flask-Based Web Application