MULTI VARIABLE ROAD FATALITY PREDICTION STUDY IN CITY AREA BY USING ANDREASSEN AND ARTIFICIAL NEURAL NETWOKS MODEL DEVELOPMENTS IN WEST JAVA, INDONESIA

Data on fatality traffic accident victims has not been fully identified in Indonesia and might have most likely been unreported. Republic of Indonesia Traffic Ordinance Number 22 Year 2009 stated that fatality data must be completed with hospitals’ data. However, the data reported by Republic of Indonesia Police has not been in accordance to the law. It causes the data to be inaccurate and unfit to be referred in the study of road safety. In many countries, researchers have been using population and motor vehicles numbers as variables to predict fatality victims’ number. Those variables are not fit with Indonesian condition that have the vastest areas, the densest population, the longest road infrastructure, and the largest numbers of motor vehicle users in ASEAN countries. The main purpose of the study was to develop better fatality prediction model in line with Indonesian condition. This was done by developing multivariable Andreassen and Artificial Neural Network models. The model was built by using population data taken from 8 cities in West Java Province, Indonesia in 2007-2010. Main results from model validation test are: (1) three variables ANN with one hidden layer prediction model was the best prediction used for to predict fatality numbers; (2) Fatality number was 122.8% bigger than that fatality data reported by Republic of Indonesia Police, that was, 956 people; (3) Andreassen prediction model was unfit to be used in Indonesia. The results were recommended as the newest prediction models to be used in the road safety study in Indonesia.