Vehicular Traffic Congestion Detection and Prediction Using Automated Supervised Learning Models

Abstract

Introduction: Vehicular traffic congestion is a major challenge in urban transportation systems, leading to increased travel time, fuel consumption, and environmental pollution. Intelligent Transportation Systems (ITS) leverage machine learning techniques to analyze traffic patterns efficiently. Automated supervised learning models enable accurate detection and short-term prediction of congestion levels. This study focuses on developing a data-driven framework for real-time traffic congestion detection and forecasting.

Objectives: The primary objective is to design an automated supervised learning framework for detecting and predicting traffic congestion levels. It aims to compare multiple classification and regression models for performance optimization. The study seeks to improve prediction accuracy using balanced and well-preprocessed datasets. Another goal is to categorize congestion into multiple levels for effective traffic management and decision-making.

Methods: Traffic data is collected from sensors, GPS devices, cameras, and external sources such as weather and events. The data undergoes preprocessing, feature extraction, and SMOTE-based balancing to enhance model performance. Supervised learning models including Random Forest, SVM, XGBoost, LightGBM, are trained and tested. Model evaluation is performed using accuracy, precision, recall, and F1-score metrics.

Results: The experimental results demonstrate that ensemble and hybrid models achieve higher detection and prediction accuracy compared to individual classifiers. The system effectively classifies congestion levels into Normal and Heavy categories. Data balancing and feature selection significantly improve model stability and performance. The framework provides reliable short-term congestion forecasts for proactive traffic control.

Conclusions: The study concludes that automated supervised learning models provide an efficient solution for real-time traffic congestion detection and prediction. Hybrid and ensemble approaches enhance accuracy and robustness in complex traffic environments. Proper preprocessing and balanced datasets are critical for optimal performance. The proposed system supports smart city initiatives by enabling intelligent and data-driven traffic management.