Comparative Assessment Between Statistical and Soft Computing Methods for Accident Severity Classification

Abstract

Globally traffic scenario has experienced a vast change over the years as volumes of traffic on urban roads have increased by leaps and bounds. This increase has also marked an increase in the number of road accidents, both fatal and non-fatal. Accident prediction has a significant role in the improvement in traffic safety, and urban traffic management. Different statistical methods and various soft computing techniques are being used for developing accident prediction models. In this study, ensemble ranking method, comprising three filter-based ranking methods such as Information Gain, Gain Ratio, and Symmetric Uncertainty, has been employed to select significant features for prediction of accident severity. Through analysis, eight significant features have been selected for accident severity prediction. These are Annual Average Daily Traffic, Road Width, Percentage of Left-Turn Vehicles, Driver’s Age, Percentage of Right-Turn Vehicles, Types of Vehicles Involved, Spot Speed, and Cause of Accidents. Multiple logistic regression (MLGR) and artificial neural network (ANN) have been applied for accident severity prediction in urban roads. This analysis concludes that ANN model gives more accuracy and minimum errors compared with the MLGR method at the time of severity classification of a vehicular accident.