基于改进密度聚类算法的交通事故地点聚类研究

交通事故特征受地域分布影响显著，本文对交通事故特征进行优化聚类研究.基于 2019年无锡市交通事故数据，调用开放地图接口地理编码解算事故地点经纬度，使用密度聚类算法对事故地点与事故原因进行密度聚类.传统的密度聚类算法依赖距离阈值和样本数阈值的准确输入，为解决这一局限，建立一种自适应搜索距离阈值和样本数阈值的密度聚类模型，并与原始聚类模型进行对比.结果表明，优化算法在参数确定上更加智能，对簇的划分更加准确，对噪声点的识别更加合理.通过机器学习中轮廓系数计算方法计算模型得分，证明了该算法在城市道路交通事故地理位置聚类中的适用性.

Traffic Accident Location Clustering Based on Improved DBSCAN Algorithm

Traffic accident characteristics are significantly affected by regional distribution. In this paper, traffic accident characteristics are clustered by the optimized density- based spatial clustering of applications with noise (DBSCAN) clustering method. The 2019 traffic accident data in Wuxi, China is used as a case study. The open map API is used to obtain the longitude and latitude of the accident location as an input for the proposed method. The traditional DBSCAN clustering algorithm normally requires accurate input of the distance threshold and sample number threshold. This paper develops the DBSCAN clustering model with an adaptive search distance threshold and sample number threshold. The comparison results of the proposed algorithm with traditional algorithm show that the optimized algorithm is more intelligent in determining parameters and more accurate in dividing clusters; the recognition of noise points is more reasonable than the traditional algorithm. The applicability of the algorithm in the geographical location clustering for urban road traffic accidents is proved by calculating the model score of the silhouette coefficient in machine learning.