Urban road traffic congestion index prediction based on a hybrid LightGBM-LSTM model

Accurate and timely prediction of the urban traffic congestion index (TCI) is crucial for implementing proactive traffic management and alleviating urban congestion. To address the limitations of single models in capturing both complex temporal dependencies and high-dimensional feature interactions, this paper proposes a novel hybrid prediction framework that synergistically integrates a Long Short-Term Memory (LSTM) network and a Light Gradient Boosting Machine (LightGBM). The model is designed to perform dual-stream learning: the LSTM module extracts medium- and long-term temporal patterns from historical TCI sequences, while the LightGBM module concurrently learns discriminative feature representations from the structured traffic data. A genetic algorithm (GA) is employed to optimize the fusion weights of the two components, constructing an adaptive and cohesive LightGBM-LSTM prediction model. The proposed framework was validated using real-world TCI data collected from three representative segments with varying congestion levels (mild, moderate, and severe) on Chengdu’s Third Ring Road, covering a period from September to October 2024. The experimental results demonstrate that the hybrid model significantly outperforms both standalone LSTM and LightGBM baselines across all test scenarios. Specifically, it achieved accuracy improvements of 4.87% and 33.06% in mildly congested sections, 26.80% and 22.32% in moderately congested sections, and 47.87% and 10.47% in severely congested sections, respectively, measured by the Mean Absolute Percentage Error (MAPE). These findings confirm that the proposed GA-optimized LightGBM-LSTM hybrid model effectively enhances TCI prediction precision and robustness by leveraging complementary strengths of sequence learning and feature engineering. The study provides a reliable and efficient analytical tool for short-term traffic state forecasting, offering valuable support for the development of data-driven and refined urban traffic management strategies.