Intelligent Urban Traffic Congestion Prediction Through Accident-Aware and Time-Dependent Traffic Analytics

Area of Science:

• Urban planning and smart city development
• Transportation engineering and traffic management
• Data science and artificial intelligence in transportation

Background:

• Rapid urban growth exacerbates traffic congestion in smart cities, leading to increased travel times, fuel consumption, pollution, and operational costs.
• Existing traffic congestion prediction models inadequately address the impact of road accidents and peak-hour conditions, crucial factors in smart city traffic dynamics.
• Both recurrent (peak-hour demand) and non-recurrent (road accidents) congestion significantly impact road network performance.

Purpose of the Study:

• To introduce a novel Historical Accident-Aware Peak-Hour GAN-GRU (APG-GRU) framework for enhanced traffic congestion prediction in smart cities.
• To integrate historical accident data with traffic flow features for more accurate congestion forecasting.
• To develop a practical tool for smart city traffic management systems.

Main Methods:

• Development of a data processing pipeline that integrates Automatic Number Plate Recognition (ANPR) traffic data with historical accident features.
• Implementation of a novel Generative Adversarial Network-GRU (GAN-GRU) architecture, termed APG-GRU, for congestion prediction.
• Experimental validation using a novel integrated dataset and external datasets (Lahore, Pakistan), comparing APG-GRU against baseline models (XGBoost, LSTM, Random Forest).

Main Results:

• The APG-GRU framework achieved superior performance on the integrated dataset, with an accuracy of 97.50%, congested precision of 91.86%, recall of 97.31%, and F1-score of 94.51%.
• APG-GRU significantly outperformed baseline models and models using only ANPR traffic data.
• External validation confirmed the robustness and generalizability of the APG-GRU framework.

Conclusions:

• The proposed APG-GRU framework effectively predicts traffic congestion by incorporating historical accident data, outperforming existing methods.
• The framework offers practical applicability for smart city traffic management through a web-based interface visualizing accident hotspots and traffic conditions.
• Accurate congestion prediction using APG-GRU can lead to optimized traffic flow, reduced travel times, and improved urban mobility.