An Innovative Approach of Fusing Weather Data for Traffic Flow Prediction

Accurate and efficient traffic prediction is essential for alleviating traffic congestion and aiding transport stakeholders in decision-making. With global climate change and the increasing frequency of extreme weather events, studies have increasingly focused on the impact of weather factors on traffic flow. However, existing research lacks a systematic approach to selecting weather characteristics and has yet to establish a unified standard. In traffic prediction, input variables are as crucial as algorithms, and the diversity of weather features across regions and climates further complicates their selection. This paper proposes a systematic process for weather variable selection, employing principal component analysis (PCA) to downscale weather features, which are then integrated into a traffic flow prediction model. Additionally, the prediction model is constructed with the traffic, weather, and combined parts to validate the proposed selection process. Standard modules commonly used in traffic prediction are utilized, as the focus of this paper is not on algorithmic improvement. Experimental results demonstrate that model performance improves after integrating the selected weather variables. This study offers a simple and practical solution for incorporating weather factors into traffic prediction models.