IDENTIFICATION OF FACTORS INFLUENCING INJURY SEVERITY OF MOTORIZED TWO WHEELER CRASHES IN PATNA (Paper 783)

Worldwide road crashes pose significant threat to social and economical life. WHO report – 2015 on road safety mentioned that the situation was more dangerous in the developing countries because of lack of proper enforcements and techniques in improving road user behavior and decreasing injury severity outcomes. In India as the urbanization is increasing, high speed road facilities have promoted the motorcyclists and other vulnerable road users (VRUs) to select these facilities. This has led to the increase in high VRUs road crashes. Lack of proper crash reporting system and complex nature of crash in heterogeneous traffic flow conditions have worsened the problem of road safety, particularly in the Patna, capital city of Bihar where crash severity in 2015 was 39.20 as compared to 29.10 of India.
The objective of this study was to identify the explanatory variables affecting the crash severity of motorized two wheelers in the Patna with the help of data mining. Two years (2014- 2015) of crash data, collected from police FIR reports, was used in the analysis. Total 17 categorical and numerical attributes such as time of a day, traffic signs, street lights, gap in medians and roadside features were used as independent variables. Roadways were divided into homogeneous segments in terms of land use pattern and vehicle mix of that area. Crash severity was divided into fatal, sever (incapacitating injury) and minor (Non- incapacitating injury). Decision tree models (J48 and random forest) were generated in the analysis of crash data because it can identify and easily explain the complex patterns associated with crash risk and do not need to specify a functional form. J48 and a Random Forest model were used using default parameters of Weka using 80% percentage split. The advantage of tree-based methods is that they are non-linear and non-parametric data mining tools for supervised classification and regression problems. They do not require a priori probabilistic knowledge about the phenomena under studying and consider conditional interactions among input data. 
Both J48 and Random forest models were effective in predicting the crash severity with classification accuracies of 54 and 59 % respectively & having kappa statistics values of 0.32 and 0.37 respectively which falls in the fair agreement range. Time of a day, no. of access/km, median openings, land use, on-street parking and street lights were found to be significant in predicting the injury severity levels in the city.