Traffic Injury Severity Prediction System using Factor Analysis based on Deep Learning

Year to year, due to the road traffic crashes which cost about 3% of the gross domestic product in most of the developed countries is recorded approximately 1.25 million of fatalities. Traffic accidents remain the major causes of death which impose a tremendous economic and emotional burden on society across the world. In the missions to tackle this universal problem, numerous studies have been conducted in an effort to identify the risk factors associated with the injuryseverity level of the crashes. So, this system is intended to explore the correspondence of factors on accident data analysis based on Mutual Information (MI) approach. The selected group of factors is used to predict the injury severity. The most recent single-vehicle crash data is obtained from the Highway Accident Information Management System, Thailand (HAIMST). In this system, mutual information that is information theory approach is used to select the maximum relevant factors for traffic injury severity prediction. By using maximum relevant factors, this system applies the convolutional neural network (CNN) deep learning method to predict the traffic injury severity levels that are no injury, minor, severe and fatal based on group factors.