DEVELOPMENT OF A METHODOLOGY FOR ROAD SAFETY FOR A NATIONAL HIGHWAY

Abstract

Road accidents are a human tragedy that result in health, environmental and social problems, and have significant impacts on national economic growth strategies. Road accidents are reflection of the lack of efficiency and safety of road transport system. The problem is acknowledged as an important health, social and financial issue globally. The World Health Organization (WHO) forecasts that road accidents will move from the ninth to the third most serious health problem in the world, within the next 10 years, if not addressed adequately. Worldwide, the number of person killed in road traffic crashes each year is estimated at almost 1.2 million, while the number injured is as high as 50 million. In India, 80,000 persons die on roads every year. The population of vehicle has increased from 1.87 to 66.83 million from 1971 to 2004 , while the road length has increased marginally from 0.58 million to 3.3 million for the same duration. Accidents have increase from 0.12 to 0.46 million. India has a road network of 3.3 million kilometers of which national highway length is just 0.07 million kilometer, however witness forty percent of total accident, of which 32.5 percent are fatal and 52 percent are serious injury accidents. Accidents burden the national economy by as much as 1% of the total GDP in addition to the losses the society and individuals suffer in terms of the pain, human loss and disability that can not be quantified. Improvement in geometry is known to improve safety from the experience of developed nations. But no such attempt has been made in India. The National Highways in India handle mixed traffic. The traffic consists of slow moving vehicles to modern passenger cars. To cater to these vehicles with extreme dynamic and static properties without compromising safety is a challenging job. One of the key question is which geometric feature be improved and to what extent. This study is an attempt to identify the key geometric and traffic parameters responsible for accidents, model their interrelationship and quantify the effect of each parameter on accident rate. Comprehensive review of literature was carried out with the purpose of providing an overall perspective of the major findings of past research work related to traffic accidents. The literature review provided information concerning the methodology of past safety research and the possible types of statistical tools that could be used. National Highways No 58 was selected as candidate for analysis. The highway serves as life line to the hilly part of the Uttaranchal and is important with respect to strategic point as it connects the capital of India (New Delhi) to the china border. The highway connects important religious destinations of the state. Rate of accidents on the highway is as high as 1.2 accidents per day. The traffic intensity on this highway is 27000 PCU on normal days and as high as 48000 PCU during tourist season. The traffic consists of31 to 34% of heavy vehicles, 8 to 14 %of non-motorized vehicles, 43 to 52 %of passengers cars and other vehicles 8 to 12 %. The highway is basically a two lane undivided but has a 53 kilometers of four lane stretch. The high rate of accidents is due to the above geometric features and traffic volume. It is because ofthis the highway is considered for this study. The scope of the study is limited to this National ii Highway 58. In the present study a total of230 Km long portion from New Delhi till Rishikesh was considered for analysis. The results of the study would be useful in framing combat policy for other national highways in the country with similar conditions. Traffic and geometry data was collected by surveys and field studies. Accident data was collected for the police stations. The detailed analysis of the data was carried out. As accidents are random events (count data) stochastic modeling technique was employed to develop accident prediction model. Stochastic family of models comprise of Poisson and Negative binomial models. Models provide relationships between observed count data and a set of explanatory variables that follow either Negative binomial or Poisson distribution. Poisson distribution suffers from the limitation of assuming mean of variable equal to variance; whereas negative binomial distribution allows variance to exceed the mean. For this study negative binomial distribution was found to be appropriate as variance was found to be more than the mean (Overdispersion of data). Separate model were developed for four lane portion and two-lane portion. Two-lane portion was further subdivided into straight segments, curve segments and unified segments (Combination of straight and curves segments). Several sets of models were developed with permutations and combinations of traffic and geometry variables. Outlier identification was done using the Cook's distance criterion. Akaike Information Criterion (AIC) was used to select final model from among the competing models along with engineering judgment. Smaller the AIC value, the better the model. Developed model were tested on the accident data of 2005. The results were found to be quite encouraging. in Modeling of accidents was also done using ANN. ANN networks were trained using the same independent variable as used in the statistical modeling to allow a comparison between the two. Multilayered Back propagation algorithm was used to train the network. 1 to 5 hidden layers with 3 to 9 neurons were used with sigmoid transfer function. The network with least mean square error was chosen as the best network. The output of ANN network was compared with statistical models output. R Square, value of the predicted against observed was calculated. It was concluded that neural network's performance was better than the statistical models. ANN proved to be an efficient tool for modeling as well as sensitivity analysis. Accident black spots were identified on the basis of statistical analysis of accident data of last five years. "Accident Rate" (AR), "Accident Fatality Rate" (AFR), "Accident per Million Vehicle Kilometers" (APMVK), and "Accident Severity lndex"(ASI),.were calculated from the past data. Any location with fatality rate more than mean fatality rate + 1.5 standers deviation offatality rate was identified as an accident black spot. Remedial measures were suggested using mathematical models based on the study results. As a part of this study a comprehensive accident database was developed for the aid of the road users, law enforcement agencies, and researchers using Geographical Information System. GIS is a computer software package to organize data with location dimension and is a versatile tool for handling georeferenced data. GIS ARC-INFO was used for map based location analysis while MS Access was used as backend engine store data. The system was found to be very handy to store accident data. IV