Exploring the Efficiency of Random Forest, Support Vector Machine, and Artificial Neural Network in Traffic Crash Data

Public safety is significantly impacted by traffic collisions, necessitating the development of reliable prediction models to enhance accident response and preventive measures. This study investigates and compares the predictive effectiveness of three prominent machine learning algorithms: Random Forest, Support Vector Machine (SVM), and Artificial Neural Network (ANN) for analyzing traffic crash data. The dataset comprises detailed information on past crashes, encompassing various variables such as month, age, gender, accident type, location, collision type, number of victims, and time period. Data from the Banepa Traffic Office of Kavrepalanchowk over the past four fiscal years is utilized. Utilizing a combination of machine learning models, including the Random Forest Classifier, SVM, and ANN, the research evaluates their performance based on the f1 score. The Artificial Neural Network exhibits slightly superior performance in predicting traffic accidents compared to the other models, achieving efficiency rates of 62.5%, 65%, and 72.5% for the Random Forest Classifier, SVM, and ANN, respectively. The findings have implications for the development of advanced decision support systems and legislative initiatives to improve road safety and mitigate the societal impact of traffic accidents. Future research is encouraged to utilize larger datasets and explore new machine learning models for identifying high-risk regions, predicting potential accident types, and identifying accident peak periods.