Real-time risk tag forecasting for safer intelligent transportation

Risk tagging is a technique of converting hazardous traffic events into numerical risk tag values using infrastructure-based edge systems. The risk tagging techniques provide advanced remote warnings to both driven and autonomous vehicles to achieve safer intelligent transportation systems. With real-time risk tagging of pedestrians and vehicles recorded at specific locations, upcoming risks can be obtained using time series forecasting methods. This paper compares five models to forecast short-term qualitative risk values based on historical time series risk tag data, including SMA, ARIMAX, LASSO, LSTM, and XGBoost. Among the five models, XGBoost provides the most accurate forecast on risk tag and has a low latency and processing time, allowing real-time forecasting. This risk warning provides drivers and autonomous vehicles with advanced warnings, thus, improving the safety of transportation systems via cooperative mode at specific locations.