The Road Safety: Utilising Machine Learning Approach for Predicting Fatality in Toll Road Accidents

Abstract

Road safety is one of the critical government transportation concerns, especially on the toll roads. With the increasing number of toll roads as part of infrastructure planning, road traffic accidents are significantly escalating. Developing a system that predicts accidents on toll roads will benefit to reduce the harm that is caused by traffic accidents. This study will propose a method for analysing toll road accidents in Indonesia using historical toll road accident data as a dataset to become a pattern to examine the frequency of accidents. This dataset consists of various parameters from three main factors that cause accidents: human, environmental, and road infrastructure factors. Machine learning technique will be mainly used to determine the most influencing factors by employing classifiers such as Logistic Regression (LR), Decision Tree (DT), Gaussian Naïve Bayes (GNB), and K-Nearest Neighbors (KNN) can construct the prediction model. Fourteen subfactors from the data were used to predict the future fatalities caused by accidents, which allowed the system to forecast the accident fatality. The results show accuracy performance on the test set with LR, DT, KNN, and GNB models, 85.3%, 79.4%, 87.1%, and 77.1%, respectively. The KNN Classifier model has the most minor error value of 0.6 compared to the other models. The study’s findings will help analyse the causal factors involved in toll road accidents and could be utilised by road authorities to employ risk control options to mitigate the ramifications.

Keywords

• road safety
• logistic regression
• decision tree
• gaussian naive bayes
• k-nearest neighbor