A Road Hotspots Identification Method Based on Natural Nearest Neighbor Clustering

• Published in: Proceedings (IEEE Conference on Intelligent Transportation Systems) pp. 553 - 557
• Main Authors: Han, Qingwen, Zhu, Yingxiang, Zeng, Lingqiu, Ye, Lei, He, Xueying, Liu, Xiaoying, Wu, Haotian, Zhu, Qingsheng
• Format: Conference Proceeding Journal Article
• Language: English
• Published: IEEE 01.09.2015
• Subjects: Algorithm design and analysis; Algorithms; cluster; Clustering; Clustering algorithms; Clusters; Conferences; distance threshold; Hot spots; hotspots identification; natural nearest neighbor; outlier retrieve; Parameter sensitivity; Road accidents; Road safety; Roads; Shape
• ISSN: 2153-0009; 2153-0017
• DOI: 10.1109/ITSC.2015.97

During the last decade, the concept of cluster, has become a popular practice in the field of road safety, mainly for the identification of worst performing areas or time slots also known as hotspots. However, current clustering methods used to identify road accident hotspots suffer from various deficiencies at both theoretical and operational level, these include parameter sensitivity, identify difficultly on arbitrary shape, and cluster number's rationality. The objective of this study is to contribute to the ongoing research effort on hotspots identification. Employing the concept of natural neighbor, a new algorithm, named distance threshold based on natural nearest neighbor (DTH3N), is proposed in this paper, striving to minimize the aforementioned deficiencies of the current approaches. Experiment results show that, comparing with existing methods, proposed algorithm presents a better performance on cluster division. Furthermore, this new method can be viewed as an intelligent decision support basis for road safety performance evaluation, in order to prioritize interventions for road safety improvement.