Risk Prediction Model of Road Accidents During Long Holiday in Thailand Using Ensemble Learning with Decision Tree Approach

Abstract

The rate of injury and death from traffic accidents during the New Year and Songkran Festival each year has high and are continuously on the increase. The researchers, therefore, has decided to study and develop a model for predicting the road accident risk during the holiday season with ensemble learning based on decision tree approach. The aim is to help reduce accidents and loss of life caused by road accidents. The dataset used in this research is traffic accidents resulting in injury and death data during the long holiday from 2008 to 2015 from hospitals across the country, accumulatively recorded by the National Institute for Emergency Medicine. Thisresearch compared the efficiency of data classification to find the best ensemble model for predicting traffic accident risk. The methods studied included Adaptive Boosting (AdaBoost), and Random Forest, and the decision tree techniques used in the experiment were J48, ID3, and CART. The results of experiment and comparisons of classification efficiency showed that the Random Forest algorithm with J48 decision tree was the most efficient model, providing an accuracy of up to 93.3%.