SIGNIFICANT SPATIAL HOTSPOT DETECTION AND ANALYSIS OF HIGH-RISK ROAD ACCIDENT ZONES USING OPTICS & HDBSCAN CLUSTERING

Abstract

Spatial Clustering is a subclass of Unsupervised Machine Learning algorithms that utilizes a specified criterion to group a set of geographically dispersed data points together. A Spatial hotspot is a confined region/space inside a geographical area which has a higher concentration of activity points than outside the hotspot region within the study area. With technological breakthroughs in spatial data collection and increased computation capabilities, several research publications presented novel methods for Spatial Hotspot detection. The relationship between Spatial Hotspots and time (temporal effect) has also been investigated by various researchers, expanding the horizon of the spatial hotspot to spatiotemporal hotspot. Spatiotemporal hotspots are a valuable tool for determining the highly vulnerable regions (e.g., high-risk crime territories, high accident-prone areas, severe disease-outbreak areas, areas more exposed to natural calamities and many more). The idea of Spatiotemporal hotspot detection gained considerable interest among researchers because of its practical utility in public health, public safety, traffic volume analysis, crime zone analysis, and other essential applications. Our research aims to organize the existing comprehensive literature into a well-organized hierarchical framework to comprehend better the various methodologies widely adopted around the world. We propose a framework that suitably categorizes the research effort of various scholars into several appropriate categories based on their differences in primary algorithmic approaches. Furthermore, we also present an extensive analysis of the widely utilized evaluation measures adopted in this research domain. Effective clustering algorithms are required in societal applications such as road safety to uncover useful patterns linked with the data. Many applications employ Spatial Scan Statistics to identify spatial clusters, however it needs users to first determine the shape of the cluster, which is ambiguous in the context of road safety and could result in unfavorable outcomes. We outlined a method for discovering statistically significant shape-invariant spatial clusters in order to identify high-risk road accident zones. The proposed approach incorporates the OPTICS and HDBSCAN clustering algorithms, as well as Cluster density and the Log Likelihood Ratio methods for evaluating the significance of spatial clusters. OPTICS and HDBSCAN are optimal for finding clusters of arbitrary shape and only require a single input parameter, making hyper parameter tuning relatively easier. We incorporated the statistical significance of clusters to eliminate spurious patterns. To illustrate the obtained results, the proposed approach is applied to the UK Road-Accidents data. We also presented a comprehensive analysis of the seasonal and temporal variations of the significant spatial hotspots discovered. Besides, we highlighted the methodology of prioritizing the identified hotspot zones based on the severity of the accidents for situations in which authorities are restricted with limited budget. We further suggest potential solutions for the probable causes of accidents in the hotspot area to guide relevant authorities to take timely and effective action.