关键词: 聚类算法, 相连的簇, 累积平均密度, 容纳因子

Abstract:

There exist two defects in the DBSCAN algorithm: input sensitivity, unable to distinguish clusters which have different density and are adjacent to one another. To solve these defects, an improved algorithm based on DBSCAN is proposed. The algorithm uses cumulative average density to determinate whether one cluster can be merged with another or not, has weakened the role of density threshold——Minpts, chooses the object with the maximal density as the beginning center object, does clustering according to the density from high to low, which is hierarchical to a degree, and hence supports clustering datasets with variable density. In the end, datasets are used to do clustering experiments. The results show that the improved algorithm has robustness of parameters to some extent, and can achieve desired effect when clustering dataset with clusters of variable density linked together.

Key words: cluster algorithm;clusters linked together;cumulative average density;accepting factor