Application of Random Forest Model for Traffic Accidents Severity Prediction in Qassim Province of Saudi Arabia

Traffic related injuries have become a threatening socio-economic dilemma in the Kingdom of Saudi Arabia (KSA), accounting for third leading source of mortalities in the country. Al-Qassim region is one of thirteen provinces in KSA which has experienced an increasing trends of traffic safety concerns over the past two decades. In literature, several studies have investigated the relationships between accidents injury severity outcomes and the contributing risk factors, however this subject is relatively under-researched in the KSA environment. The primary aim of this research work is to model injury severity of traffic accidents based on Random Forest (RF) model). The target variables (accident) injury severity had binary outcomes, i.e., fatal or nonfatal. Further, to assess the influence and relative role of input accident predictors (humans, vehicles, roadway, and weather characteristics), the study also examined applicability of RF sensitivity analysis technique. Different statistical indices including accuracy, precision, recall, F-1 scores, and area under the curve (AUC) were used to test the injury severity classification and predictive performance of proposed RF model. Based on empirical findings it was revealed that the RF model achieved an overall classification accuracy of about 78%. Model evaluation based on other metrics also demonstrated acceptable performance of RF for injury severity classification. According to RF feature ranking analysis it was evident that risk factors including accident type, weather conditions, illumination (lighting) conditions, and time of the day (TOD) were the leading severity risk factors. The findings of this research can provide the local authorities better insights for proactive safety interventions to improve traffic safety.