Comparative Study on Spatial Clustering Methods for Identifying Traffic Accident Hotspots

Traffic accidents in an urban road network are inevitable as a result claims and disputes arise among different road users. It is imperative to estimate the likelihood of traffic accidents resulting from different factors that contribute to loss of life, property and health of road users. There is a pressing need to reduce traffic accidents by identifying the location of accident hotspots using suitable analysis methods and examining them which is essential for the safety of road users. In this research traffic accident hotspots are identified using two spatial clustering analysis methods namely Getis-Ord Gi* and Nearest Neighborhood Hierarchy (NNH). These methods are compared and evaluated using the Prediction Accuracy Index (PAI) for their degree of accuracy. In this study, a cumulative traffic accident data of Hyderabad city of Telangana state over four years is researched upon and considered. Getis-Ord Gi* analysis measures the concentration ratio based on Z score identified as high (positive Z-values) and low values (negative Z-values). NNH analysis is another spatial clustering method which displays hotspot regions in the form of Convex hulls and Ellipses. The choice of the above two clustering methods represents the significance of the precision required. The findings of the study reveal that NNH method performed better compared to Getis-Ord Gi* method in its ability to detect hotspots. The above research methodology can be performed to any size of road network area globally having relevant accident data for the identification of hotspots for reducing the traffic accidents.