Predicting Road Traffic Accident Using Genetic Algorithm

Abstract

Predicting traffic accidents is a difficult task for everyone involved in motor vehicle trafficsince it has the potential to save lives. Predicting the severity of traffic accidents has been attemptedwith a variety of classification methods, Artificial neural networks, support vector machines, decisiontrees, and more are included in this category. It is true that neural networks and support vector machineshave a huge drawback when it comes to human interpretation (SVMs), whereas low accuracy is a majordrawback of classic deep learning techniques like C4.5, ID3, and CART. This paper proposes analternative to classic DTs based on user preferences in order to address the shortcomings of currentmethods for predicting traffic accidents. In order to optimise and uncover rules based on therecommended method's support, confidence, and comprehension criteria, we created a new geneticalgorithm. Users, such as traffic cops, road and transportation engineers, will be able to apply theirexpertise while balancing all of the conflicting objectives using the proposed technique. A five-yearperiod (2008–2013) of Tehran Province, Iran, traffic accident data was utilised to evaluate therecommended approach. When it comes to classification measures like accuracy and rule performancemetrics, this strategy outperforms classification methods like ANN, SVM and conventional DTS.