A Data-Driven Approach to Intelligent Traffic Control and Congestion Classification

Traffic congestion lengthens journeys, increases fuel use, and delays emergency vehicles. Fixed-time signals cannot adjust to demand that changes through the day, while fully adaptive systems need a lot of sensing hardware. This paper describes a prototype web-based Smart Traffic Management System that classifies congestion as LOW, MEDIUM, or HIGH and suggests a green time for each signal. The system uses a React dashboard, a Node.js API, Supabase (PostgreSQL) for storage, and a Random Forest classifier. The model is trained on a simulated dataset of 50,000 rows covering 20 roads and signals in Bhopal. The inputs are the number of vehicles, average speed, previous traffic density, hour, day of week, weather, and event type. On a random 80/20 split the deployed model reached an accuracy of 88.22% and a macro F1 score of 88.51%. Logistic regression and a 200-tree Random Forest gave almost the same results, which shows that the simulated classes can be separated using only a few variables. Because the data is simulated and the split is random, the results show that the pipeline works end to end, but they do not show any benefit on real roads. The system has not yet been tested with real sensor data.

Keywords - Smart Traffic Management, Congestion Classification, Random Forest, Signal Timing, Emergency Vehicle Priority, Simulated Data