Novel weather severity classifications based on speed reductions utilizing machine learning techniques

Inclement weather significantly impacts road safety and mobility, yet current weather severity classifications primarily rely on fixed, precipitation-based thresholds, which limits their effectiveness under complex conditions. This research proposes a novel weather severity classification leveraging weather-induced speed reductions and machine learning techniques. By integrating Road Weather Information Systems (RWIS), INRIX probe vehicle speed data, and volume data from permanent count stations (PCS), this method captures diverse meteorological factors, including visibility, wind speed, and road surface conditions. K-means clustering is utilised to classify severity based on observed speed reductions, and predictive models using multinomial logistic regression, decision trees, random forests, XGBoost, and multilayer perceptron neural networks are developed and validated. Results show higher classification accuracy using ensemble methods, especially random forests and XGBoost. By implementing this scalable, data-driven approach, transportation agencies can enhance proactive traffic management strategies, improve alert timeliness, reduce crash risk, and maintain mobility during adverse weather conditions.