Comprehensive density-based cluster analysis.

摘要

Clustering is one of the primary techniques in data mining. Clustering in data mining is a discovery process that partitions the data set into groups such that data points within a group have high similarity in comparison to one another but are very dissimilar to points in other groups. Clustering has the following several challenges: (1) clusters with arbitrary shapes, (2) minimal domain knowledge to determine the input parameters, and (3) scalability for large data sets.; Density-based clustering has been recognized as a powerful approach for discovering arbitrary-shape clusters. However, the other two challenges still remain in most existing clustering algorithms. The goal of this dissertation is to explore comprehensive clustering methods to meet the current challenges. We conquer the second challenge by two attempts: (1) to reduce input parameters in the density-based partitioning algorithm and (2) to eliminate input parameters by means of hierarchical clustering. We meet the third challenge by improving the speed by means of the vertical P-tree structures and pruning techniques.