Abstract:

Clustering is a critical analytical tool in data mining. An improved densitybased clustering algorithm concerning local center object is proposed based on the classical DBSCAN algorithm. In this algorithm, the density of each object is measured by its knearest neighbors. The object which has the local maximal density is regarded as the center object. The data set is divided into different districts. The cluster expands from the center object until the edge decided by the scalefactor of density. To improve the quality of clustering, the candidate core object is proposed in the algorithm. Core objects can be recognized from candidate core objects, so that current clustering is expanded. The experimental results show that the proposed algorithm is considerably effective in reducing the sensitivity of the parameters, improving the clustering results with maldistribution of data sets and enhancing the clustering accuracy.

Key words: clustering;DBSCAN;density;local center object;k-nearest neighbors