A Lightweight Machine Learning Model for Predicting Traffic Congestion in Urban Areas

Urban traffic congestion is a critical challenge in modern cities, impacting economic productivity, environmental sustainability, and quality of life. This paper proposes a lightweight machine learning model designed to predict traffic congestion in urban areas with high accuracy and computational efficiency. Leveraging real-time data from IoT sensors, GPS devices, and historical traffic patterns, the model employs an optimized ensemble learning approach to minimize latency and resource consumption while maintaining robust predictive performance. Key features include dynamic feature selection, adaptive thresholding for congestion classification, and a streamlined architecture suitable for deployment on edge devices. The model is evaluated using datasets from three major metropolitan areas, demonstrating superior performance compared to traditional methods like ARIMA and deep learning-based solutions in terms of speed (up to ۴۰% faster inference) and accuracy (F۱-score of ۰.۹۲). Practical applications for urban planners and traffic management systems are discussed, including real-time alerts, route optimization, and policy interventions. The paper also addresses scalability challenges and proposes a federated learning framework to enhance privacy and reduce data centralization risks. By balancing predictive power with operational efficiency, this work contributes to the development of accessible, real-world solutions for smart city infrastructure.

کلیدواژه ها:

Traffic congestion prediction ، lightweight machine learning ، urban mobility ، IoT ، edge computing ، smart cities