Weather-Informed Traffic Solutions: LSTM-Based Congestion Prediction in Islamabad

Abstract

Addressing the persistent issue of traffic congestion in Pakistan, particularly in Islamabad, remains a critical challenge impacting both productivity and quality of life. Past efforts have focused on congestion prediction using parameters such as road conditions, traffic speed, and holidays, employing algorithms like Random Forest, DBSCAN, and GRU. While past research has shown promise, a significant research gap exists concerning integrating weather data specific to Islamabad's roads. This study proposes a novel approach to predict traffic congestion levels by combining 2022 Islamabad Roads Traffic Data obtained through Google APIs with weather data from the Open-Meteo website. The roads are divided into two junctions, each comprising "Cross" and "Reach" sections, with traffic flow classified into three categories: low congestion (Class 0), mild congestion (Class 1), and congested (Class 2). The predictive modelling primarily relies on Long Short-Term Memory (LSTM) networks. Two comprehensive studies were conducted: one incorporated traffic and weather data, and the other utilised only traffic data. The evaluation metrics encompassed F1 scores, precision, recall, and overall accuracy for each junction and congestion class. Notably, integrating weather data yielded superior results across all metrics, indicating a more accurate prediction of congestion levels. For instance, at Junction 1 Cross, incorporating traffic and weather data achieved an overall accuracy of 0.98, compared to 0.78 when using traffic data alone. Similarly, at Junction 1 Reach, the combined dataset yielded an overall accuracy of 0.96, outperforming the 0.88 accuracy with traffic data alone. Junction 2 Reach also demonstrated improved accuracy with the combined dataset, achieving 0.94 compared to 0.75 with traffic data alone. Future research avenues include incorporating additional parameters such as road conditions, public holidays, signals, traffic flow, density, and speed. These enhancements could further refine the predictive capabilities, ultimately contributing to more effective traffic management strategies in Islamabad and beyond.