一种多属性的时空数据聚类算法分析研究

时空聚类(spatial-temporal density based spatial clustering of applications with noise,ST-DBSCAN)算法只能处理固定属性的时空数据,且其人为设定阈值的方法具有较大随机性会导致聚类结果不理想.基于ST-DBSCAN算法存在的不足,提出了一种改进的多属性时空聚类算法.改进后的新算法采用绘制时空对象距离频数柱状图来设定自适应阈值,通过引入Gower相似系数、Dice相似系数与欧几里德距离来构建多属性相似度模型,计算多个事务对象之间属性特征的相似度大小,从而将ST-DBSCAN时空聚类算法扩展到更多属性的时空数据聚类分析中.以北京市计算机行业职位招聘信息数据进行实验仿真,实验结果表明,新提出的阈值设定方法可以有效识别部分低密度簇,提高聚类的准确性和有效性;改进后的算法具有较好的普适性与包容性,能对多属性的时空数据进行很好的聚类分析.

Research on a multi-attribute spatial-temporal data clustering algorithm

Spatial-temporal density based spatial clustering of applications with noise (ST-DBSCAN) can only deal with the spatial-temporal data analysis of fixed attributes, and due to large randomness while setting the threshold artificially, the clustering result is not ideal. Based on the shortcomings of ST-DBSCAN, we propose an improved multi-attribute spatial-temporal clustering algorithm. The algorithm calculates the similarity degree of attribute features among multiple transaction objects by constructing the hybrid attribute similarity model, which introduces Gower similarity coefficient, Dice similarity coefficient and Euclidean distance. It extends the ST-DBSCAN spatial-temporal clustering algorithm to spatial-temporal data clustering analysis of more attributes; the improved algorithm uses the spatiotemporal distance frequency histogram to set the adaptive threshold. Experiments were carried out with the recruitment information of Beijing computer industry as experimental data. And the results illustrate that the improved algorithm has better universality and inclusiveness, and can cluster multi-attribute spatial-temporal data well. At the same time, the proposed threshold setting method can effectively identify some low-density clusters and improve the accuracy and effectiveness of clustering.