NEURAL NETWORK PARADIGMS IN CRASH MODELING ON NON URBAN HIGHWAYS IN INDIA

Abstract

Engineers and researchers in the transportation discipline have tried to build safe roads following appropriate design standards, but traffic accidents are unavoidable. Patterns involved in crashes could be detected if accurate prediction models capable of automatic prediction of various traffic accidents are developed. These accident patterns can be useful to develop traffic safety control policies. To obtain the greatest possible accident reduction effects with limited budgetary resources, it is important that measures are based on scientific and objective surveys of the causes of accidents and severity of injuries. A number of explanatory variables related to traffic and road geometry that contributes to accident occurrence can be identified and to develop accident prediction models. The accident prediction models reported in literature largely employ the fixed parameter modeling approach, where the magnitude of influence of an explanatory variable is considered to be fixed for any observation in the population. The mixed traffic on Indian multilane highways comes with a lot of variability within, ranging from difference in vehicle types. This could result in variability in the effect of explanatory variables on accidents across locations. The study aims to evaluate Road Safety of a section on four-lane National Highway (NH)-58 located in the state of Uttarakhand, India. Artificial Neural Networks (ANNs) models with different training functions were employed to develop road traffic crash prediction system. ANN models with different training functions further with different number of layers and hidden neurons were trained and analysed. A total of 275 dataset were randomly divided for training, validation and testing. Results show Schawarz’s Bayesian Criterion (SBC) for ANN3 and ANN7 models were -2.135 and 1.378 respectively and the calculated model Chi Square value (38.60 for ANN3 Andamp; 23.971 for ANN7) were also lesser than the critical chi-square value(295.35) revealing the model fitted the data precisely. The results also showed that percentage of trucks in the traffic stream, spot speed 2 increased the likelihood of accident occurrence whereas adequate carriageway, shoulder and median widths decreases the occurrence of crashes.