Traffic accident severity prediction using a novel multi-objective genetic algorithm

ABSTRACT Prediction of traffic accident severity is a motor vehicle traffic challenge due to its impact on saving human lives. There are several researches in the literature to predict traffic accident severity based on artificial neural networks (ANNs), support vector machines (SVMs), decision trees (DTs) and other classification methods. In fact, the main disadvantage of ANNs and SVMs is lack of interpretation for human and the main disadvantage of classical DTs such as C4.5, ID3 and CART is their low accuracy. To address these drawbacks, in this paper we propose a novel rule-based method to predict traffic accident severity according to user's preferences instead of conventional DTs. In the proposed method, we customised a multi-objective genetic algorithm, i.e. Non-Dominated Sorting Genetic Algorithm (NSGA-II), to optimise and identify rules according to Support, Confidence and Comprehensibility metrics. The goal of the proposed method is providing facilities to make use of the knowledge of users, including traffic police, roads and transportation engineers and trade-off among all the conflicting objectives. The proposed method is evaluated by a traffic accident data set including 14211 accidents in rural and urban roads in Tehran Province of Iran for a period of 5 years (2008–2013). The evaluation results revealed that the proposed method outperforms the classification methods such as ANN, SVM, and conventional DTs according to classification metrics like accuracy (88.2%), and performance metrics of rules like support and confidence (0.79 and 0.74, respectively).