Predicting Road Crash Severity Using Classifier Models and Crash Hotspots

The rapid increase in traffic volume on urban roads, over time, has altered the global scenario. Additionally, it increased number of road crashes, some which are severe and fatal nature. identification hazardous roadway sections using spatial pattern analysis crashes recognition primary contributing factors may assist reducing severity (R.T.C.s). For crash prediction, along with patterns, various machine learning models used, relations R.T.C.s neighboring areas evaluated. In this study, tree-based ensemble (gradient boosting random forest) a logistic regression model compared for prediction R.T.C. severity. Sample data Al-Ahsa, eastern province Saudi Arabia, were obtained from 2016 to 2018. Random forest (R.F.) identifies significant features strongly correlated R.T.C.s. findings showed that cause type collision most crucial elements affecting injuries crashes. Furthermore, target-specific interpretation results distracted driving, speeding, sudden lane changes significantly contributed method surpassed other terms injury severity, individual class accuracies, collective accuracy when k-fold (k = 10) based performance metrics. addition taking into account approach, study also included autocorrelation G.I.S. identifying hotspots, Getis Ord Gi* statistics devised locate cluster zones high- low-severity demonstrated research area’s dependence was very strong, patterns clustered distance threshold 500 m. analysis’s approaches, Gi*, index, accident incidents according Moran’s I, found be successful way locating rating hotspots techniques used could applied large-scale while providing useful tool policymakers looking improve safety.