Feature Selection in Machine Learning Models for Road Accident Severity

Traffic accidents are a major cause of serious injuries and deaths around the world. Building predictive models from traffic data can give insights that help authorities improve road safety. Feature selection is an important step in building effective machine learning models. Feature selection methods are used to determine features that are relevant to classification task. The chosen feature selection method can affect the performance of machine learning models. In this paper, a real dataset of traffic accidents in Saudi Arabia is used to model accident severity. Classification models are built using single and ensemble classification algorithms. In addition, we evaluate the performance of developed models to which feature selection is applied. Two feature selection methods are used in this study: information gain, which is a filter-based feature selection method, and a genetic algorithm, which is a wrapper-based method. Experimental results show that better classification performance is obtained with genetic algorithm feature selection. In particular, ID3 and naive Bayes classifiers have improved results with genetic algorithm feature selection.