Severity Prediction with Machine Learning Methods

Abstract

Traffic, which became a part of our lives with the spread of cars, brought accidents and death with it. Safety and accident issues are a global problem in the world. This study aims to establish models to predict the accident severity levels of traffic accident injury records for possible accidents by using some data mining classification methods. The dataset used for this work is named Stats19, which has the traffic accident data from 2010 to 2012 in United Kingdom (UK), and it is collected by the UK government data service. The dataset was classified into three accident severity categories, which are fatal, serious, and slight. Classification algorithms use prior knowledge as training data to classify data objects into groups, which is good for us to work with. The models that we used are Multi-layer Perceptron (MLP), Decision Tree classifier, and Random Forest classifier and Naive Bayes classifier. The data extracted from the dataset will make sense to compare and predict a level degree. The tested classification algorithms come up with the results, the decision tree algorithm with an accuracy of 80.74%, the random forest classifier with an accuracy of 85.19%, the Naive Bayes algorithm with an accuracy of 83.40% and the MLP model with an accuracy of 86.67%. These factors of accidents can be important for estimating accident costs, increasing safety, and determining a strategy. Although it is not possible to stop accidents, it aims to reduce injury levels. This study is written in python programming language using Spyder integrated development environment (IDE).