A HYBRID SUPERVISED LEARNING APPROACH FOR RELIABLE VEHICULAR CONGESTION PREDICTION IN SMART CITIES

Abstract

Introduction: Urban vehicular congestion continues to pose a serious problem for transportation systems, causing delays, excessive fuel consumption, and greater environmental impact. Intelligent Transportation Systems (ITS) apply machine learning algorithms to systematically analyze and model traffic behavior. By using automated supervised learning techniques, congestion levels can be identified accurately and predicted in the short term. This study proposes a data-centric approach to enable efficient real-time traffic congestion detection and forecasting.

Objectives: The main objective of this research is to develop an automated supervised learning system for traffic congestion detection and prediction. It focuses on evaluating various classification and regression algorithms to determine the most effective model. The study also aims to enhance predictive performance through proper data preprocessing and balancing techniques. Additionally, it intends to classify congestion into different levels to support efficient traffic control and informed decision-making.

Methods: Traffic information is gathered from various sources, including road sensors, GPS systems, surveillance cameras, and external factors such as weather conditions and public events. The collected data is then processed through cleaning, feature engineering, and SMOTE-based data balancing to improve model effectiveness. Several supervised learning algorithms—such as Random Forest, Support Vector Machine (SVM), XGBoost, and LightGBM—are trained and validated on the prepared dataset. Performance assessment is conducted using evaluation metrics including accuracy, precision, recall, and F1-score.

Results: The experimental findings indicate that ensemble and hybrid approaches outperform standalone classifiers in both detection and prediction accuracy. The proposed system successfully categorizes traffic conditions into multi level like Low, Normal,High and Heavy congestion levels. Techniques such as data balancing and feature selection contribute notably to improved model robustness and overall performance. The developed framework delivers dependable short-term congestion predictions, enabling proactive and efficient traffic management.

Conclusions: The research concludes that automated supervised learning techniques offer an effective approach for real-time detection and prediction of traffic congestion. Ensemble and hybrid models improve both accuracy and resilience when dealing with complex and dynamic traffic conditions. Comprehensive data preprocessing and balanced datasets play a vital role in achieving optimal predictive performance. Overall, the proposed framework contributes to smart city development by facilitating intelligent, data-driven traffic management strategies.