A comparative analysis of heterogeneity in road accident data using data mining techniques

Road accidents are one of the most imperative factors that affect the untimely death among people and economic loss of public and private property. Road safety is a term associated with the planning and implementing certain strategy to overcome the road and traffic accidents. Road accident data analysis is a very important means to identify various factors associated with road accidents and can help in reducing the accident rate. The heterogeneity of road accident data is a big challenge in road safety analysis. In this study, we are making use of latent class clustering (LCC) and k-modes clustering technique on a new road accident data from Haridwar, Uttarakhand, India. The main focus to use both the techniques is to identify which technique performs better. Initially, we applied LCC and k-modes clutering technique on road accident data to form different clusters. Further, Frequent Pattern (FP) growth technique is applied on the clusters formed and entire data set (EDS). The rules generated for each clusters do not prove any cluster analysis technique superior over other. However, it is certain that both techniques are well suited to remove heterogeneity of road accident data. The rules generated for each cluster and EDS proves that heterogeneity exists in the entire data set and clustering prior to analysis certainly reduces heterogeneity from the data set and provides better solutions. The rules for Haridwar district reveals some important information which can used to develop policies to prevent and overcome the accident rate.