Impact of Rainfall Data Quality on Traffic Prediction Accuracy Using Neural Networks

Abstract

Heavy rainfall is a key factor not only in triggering natural disasters such as flash floods and landslides, but also in worsening traffic congestion, which is a significant economic and social issue. Accurate detection of rainfall behaviour can enhance the development of effective traffic prediction tools. This study aims to investigate the impact of using rainfall input data with different quality on the performance of traffic prediction using Neural Networks (NN) techniques. The challenge of this research is utilizing weather radar data, which provides high-resolution, pixel-based rainfall data at a 10-minute temporal resolution and comparing it with point-based rain gauges measurements. In this study, Phahonyothin Road, near Kasetsart University, was selected as the study area for traffic prediction. NN techniques were employed to develop a predictive model. Radar data from Sattahip radar and traffic data from Google Traffic, covering the study area, were collected and analyzed to generate the input dataset for the NN model. The results indicate that integrating radar rainfall data with traffic data significantly improves the model’s accuracy in predicting traffic conditions 30 minutes in advance, compared to models using only rain gauges rainfall data or past traffic conditions. This highlights the potential of high-resolution radar data in enhancing real-time traffic forecasting.