International Journal of Scientific Research and Engineering Development

Rapid urbanisation and the sustained growth in vehicle ownership have made traffic congestion one of the most persistent problems facing modern road networks, contributing to fuel wastage, lost productivity, and elevated emissions. Past trafficmanagement approaches, which rely on signals and monitoring traffic. In paper present to explain and empirical evaluation the CNN based pipeline for classify the traffic congestion on images. A reproducible, seed-controlled synthetic traffic-image generator was built to remove dependence on an external labelled dataset, producing 10,000 topdown 64×64×3 road images distributed across three congestion levels Low, Medium, and High (Jam) based on vehicle count. A four-block CNN (Conv2D + Batch Normalization + Max Pooling, with 32/64/128/256 filters) followed by a two-layer dense classification Reduce LRO n Plateau callbacks. On an untouched 2,000-image test partition, the model having high accuracy in testing of 93.10%, score of F’s is average of 0.93, or confusion matrix in which all residual misclassifications were confined to adjacent congestion classes, indicating that the network learned an ordered, density-aware representation despite being trained as a plain categorical classifier. A supplementary benchmarking study additionally compared the CNN architecture against five classical machinelearning baselines Decision Tree, Logistic Regression, and other type of architecture like RF, ANN,, SVM on a structured traffic-image dataset, where the CNN again achieved the highest accuracy (96.82%), recall, precision, F1-score, ROC-AUC of all models evaluated. These results, together with a comparison against CNN-based traffic-sign recognition and flow-prediction literature, demonstrate that a lightweight, self-contained CNN can classify road-traffic congestion with high accuracy and structured, interpretable error behaviour, without requiring temporal (LSTM/GRU) components.

Ritu, Amandeep, Anikit, Dharmender Kumar, "Traffic System Management Using Convolutional Neural Networks: A Reproducible CNN Pipeline for Road-Traffic Congestion Classification" International Journal of Scientific Research and Engineering Development, V9(4): Page(914-924) July-August 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.