ANALYSIS OF TRAFFIC ACCIDENT DATA OF DHAKA-BANGLABANDHA NATIONAL HIGHWAY IN BANGLADESH USING DATA MINING TECHNIQUES

Accidents happen unpredictably and involuntarily that usually cause damage or injury or fatalities. Data mining has proven to be a dependable technique for analyzing traffic accidents and results. In recent advances, data mining has taken great steps in extracting latent models and new data from large datasets often ignored by conventional statistical methods. In this study, we have identified more accurate and useful patterns that appear in traffic accident data using most decision tree induction algorithms. These patterns exercised to minimize the number of accidents or diminished the severity of an accident. As part of this research work, details of accidents occurred in Dhaka-Banglabandha (N5) National Highway of Bangladesh in the year 2007-2015 were collected from ARI (Accident Research Institute), BUET (Bangladesh University of Engineering and Technology), Bangladesh. We have analyzed 1283 traffic accident incidents through 28 features generated from the collected data. Also, we have found a few classifiers to make use of various decision tree algorithms. In this experimental analysis, we have measured the performance of 10 decision tree classifiers and also compared the error rate of the following classifiers and found J48 is the best classifier. Finally, we have extracted rules for the trees, which can predict the type of accident based on the identified attribute values. The results help to point out the factors that affect traffic accidents and analyze the more precise causes or situations of occurred accidents.