Normalized Traffic Features Using Graph Signal Processing for Traffic Flow Prediction

Traffic forecasting is essential for the intelligent transportation system and is applied to managing traffic networks. Recently, there has been a growing interest in incorporating spatial and temporal variables into the methodology of traffic forecasting. The primary emphasis of modern research lies in the complex architecture of graph neural networks. In contrast, preprocessing data for the input is also essential. Especially traffic data, such as traffic flows, have high nonlinearity and complex patterns. The traffic graph signal is attributes that are associated with the nodes. In this paper, the step of preprocessing data utilized graph signal processing (GSP) to filter noise from traffic signal data before the traffic flow data was applied for prediction tasks. The adjusted parameter for filtering noise was used. In this paper, the parameter was set from 0.1 to 10. The result shows that the increasing parameter filter of the graph data was smoothed filtered. Moreover, the smooth data is affected by the time horizons. For the increasing time horizon, we can use the large parameter and get the prediction error smaller than the shorter time horizon.