A Binary Particle Swarm Optimization Based Hybrid Feature Selection Method for Accident Severity Prediction

According to recent reports, road traffic injuries are the leading cause of death among children and young adults. Various systems and strategies have been designed to reduce accident severity. With the development of data mining tools, the use of big traffic data and machine learning techniques holds potential for implementing effective road safety strategies. Using a dataset collected from Addis Ababa, Ethiopia, our research introduces an innovative severity prediction system integrating a hybrid Feature Selection (FS) approach, named OWABPSO, with machine learning algorithms. The OWABPSO approach combines a One-Way-ANOVA-based filter method with a Binary Particle Swarm Optimization-based wrapper method. Six algorithms, including K-Nearest Neighbors, Random Forest, Decision Tree, Light Gradient Boosting Machine, Artificial Neural Network, and Extreme Gradient Boosting, are proposed for severity prediction. Experimental outcomes of this work demonstrate that, compared to state-of-the-art methods, by combining our FS approach with Decision Tree-based classifiers, we achieved competitive results. Our study presents an effective integration of FS approaches in predicting accident severity levels, thus contributing to advanced road safety strategies.