Prediction of accident severity for overtaking maneuvers on two-lane roads using Decision Tree, Random Forest, and K-Nearest Neighbors models

Abstract

Overtaking maneuvers are both dangerous and difficult to perform on two-lane roads; because of the complexity of this maneuver, there is a pressing need to intensify scientific research to find real solutions, therefore, to avert accidents and mitigate the significant mortality and economic expenses involved. This research investigates the use of three machine learning algorithms to predict the severity of accidents depending on the different factors that contribute to accidents, including Random Forest, Decision Tree, and K-Nearest Neighbors. Such variables as road surface, weather, driver gender, type of fuel, etc. are considered to train the prediction models with the accident-related datasets. In order to find out serious predictors of the severity of the accident, the performance of each algorithm is evaluated through a number of measures, such as precision, accuracy, recall, and F1 score. Regarding predictive accuracy, the findings demonstrate that RF is better than the rest of the other models, although there is a minor difference between the results of Decision Tree and KNN. The current research is a credible resource to be utilized by the transportation authorities to mitigate the effects of traffic accidents by showing the effectiveness of the machine learning techniques in predicting the main factors that have the highest impact on the accidents.