Data integration and clustering for real time crash prediction

Abstract

Cities are facing increasingly challenges with the projected population growth and the resulting increased urban travel demand. Road safety is a major issue in urban planning. Much of the empirical research on road safety and determining the probability of accidents has focused on the accident events. While human error and mechanical failure are common causes of road traffic accidents, the importance of spatial factors has been underestimated. Classifications of road accident are generally based on the available data associated with the accident itself such as time of day, type of injury, type of vehicle, etc. This can often limit the understanding of the complexities of road accidents which can be the outcome of a number of environmental, social and economic factors neglected by the standard accident data collection. This study is based on the assumption that road accidents occurring in similar areas are spatially dependent because of the increased density of accidents in the specific area. This dependence is argued to be the result of a shared common cause(s) between the accidents, albeit of varying intensity. Analysis of road accidents, particularly the spatial patterns of accidents requires further attention. This study aims to highlight some of the gaps in the research with particular attention to spatial classification of road accidents. This paper proposes an approach towards developing a comprehensive classifier system which utilizes historical as well as real-time data for any given road and outputs the probability of accident on any given time on that road. This classifier system is built by considering the exclusive spatial attributes of every road to determine the severity and importance of these attribute data on frequent accidents.