Urban Traffic Flow Prediction via Multi-Model Classifier Ensemble

Urban traffic congestion poses a persistent challenge in rapidly developing cities. This paper proposes a heterogeneous machine learning ensemble integrating Random Forest (RF), Decision Tree (DT), and K-Nearest Neighbors (KNN) classifiers to predict traffic flow with high accuracy. Trained on a two-month dataset from Kabul's Abdul-haq Crossroad with 5,952 temporal records across four vehicle types and six traffic features, the model uses weighted majority voting to fuse predictions. It achieves 99.53% accuracy-outperforming standalone classifiers by up to 1.38 %. Evaluation against four baselines (RF, KNN, DT, Naïve Bayes) confirms superiority across precision (99.62%), recall (99.35%), and F1-score (99.41 %), with a 0.47 % misclassification rate across congestion levels. ROC-AUC scores (0.996-0.998) reflect near-perfect separability. The ensemble leverages RF's feature resilience, DT's interpretability, and KNN's temporal awareness, offering a robust solution for adaptive traffic management systems. Future work includes real-time IoT integration and spatiotemporal graph neural extensions.