Hybrid K-Means-LSTM Model for Traffic Volume Prediction on Pekanbaru Arterial Roads

Abstract

Traffic congestion remains a critical issue in rapidly growing urban areas, including Pekanbaru City, Indonesia. Accurate traffic volume prediction is essential to support effective traffic management and proactive decision-making. This study proposes a hybrid clustering–sequence model by integrating K-Means Clustering and Long Short-Term Memory (LSTM) to improve urban traffic volume prediction on arterial roads. Hourly traffic volume data collected from Jalan Jenderal Sudirman were used as the primary indicator of congestion due to their strong relationship with traffic density and road capacity utilization. The research framework consists of data preprocessing, traffic pattern clustering using K-Means, and time-series prediction using LSTM, where cluster labels are incorporated as additional input features through one-hot encoding. The optimal number of clusters was determined using the Elbow and Silhouette methods, while prediction performance was evaluated using error-based metrics. The experimental results demonstrate that the hybrid K-Means-LSTM model outperforms the standalone LSTM model, particularly during peak traffic periods, by reducing prediction error and improving temporal pattern recognition. Furthermore, the clustering results provide meaningful interpretations of traffic conditions, categorized into low, medium, and high congestion levels. These findings indicate that integrating traffic pattern segmentation into sequence learning enhances both predictive accuracy and model interpretability. The proposed approach offers practical insights for data-driven urban traffic management and supports the development of intelligent transportation systems.