Theory And Practice Of Soft-subspace Cluster Based On Artificial Bee Colony Algorithm

Clustering algorithm divides a data set into several clusters by some potential andhidden features. The data objects in the same cluster are similar to each other and asdifferent as possible with the other clusters.Clustering high dimensional data is new problem in clustering field. Subspaceclustering is an effective approach for processing large-scale and high dimensional data sets.Subspace clustering has many kinds of methods. This thesis focuses on a special subspaceclustering methods called soft subspace clustering.Artificial Bee Colony (ABC) algorithm is an optimization algorithm. The algorithmuses global optimization method to search optimal solution. The algorithm has fastconvergence speed and wide application fields.This thesis proposes a new clustering algorithm by combining soft subspace clusteringalgorithm and artificial bee colony algorithm.Through improving the objective function and search policy of soft subspace clustering,a new algorithm called ABCSC (soft subspace clustering based ABC) is proposed. Theinter-cluster dispersing matrix of objective function in soft subspace clustering is weighted.Global optimal search of ABC in introduced as search policy. Experiments show thatABCSC algorithm has better performance on high dimensional data by improving objectivefunction of soft subspace clustering and introducing ABC algorithm.Experiments use standard data sets from UCI. The accuracy rate, RI (Rand Index) andNMI (Normalized Mutual Information) are calculated and compared. Through analysis ofexperiment results, the performance of the new algorithm, weakness, and possible furtherimproving is discussed.