The Development of Multivariable Road Fatality Predictive Model for Provinces Region in Indonesia
Supratman Agus · 2015
This study investigated the development of multivariable road fatality predictive model for provinces region in Indonesia. It was carried out based on a consideration that Indonesia has still used Smeed (1949) and Andreassen (1985) models as road fatality predictive models, despite the fact that two variables in these models including population and motor vehicles numbers are not suitable with infrastructure and facilities characteristics of Indonesian road transportation. The sample used in this study was West Java Province as the representative of 34 other provinces in Indonesia with the highest numbers of populations and motor vehicles, the longest road infrastructure in Java Island and the third rank in the number of road fatality in Java as the most densely populated island in Indonesia. This study aimed to explore the use of Smeed/Andreassen variables, to develop these variables into multivariable and to develop Artificial Neural Netwok (ANN) model to forecast fatality based on road transportation infrastructure characteristics in the research site, West Java Province, Indonesia. The findings showed that (1) three-variable ANN model with two hidden layers (ANN3-2HL*S) is the best fatality predictive model in Indonesia province areas; (2) ANN3-2HL*S model is able to reveal 97.5% of data reported by RI National Police Department and hospitals survey fatality result; (3) Andreassen predictive model is unsuitable for Indonesia. This research recommends two following aspects: (1) ANN3-2HL*S predictive model can be used in Indonesia provinces with similar research site areas characteristics, and (2) ANN3-2HL*S can be used for completing RI National Police Department fatality data.