Enhancing Semisupervised Clustering By Combining Rough Set Based Feature Reduction and Particle Swarm Optimization

摘要

In recent years, many applications often face the problem of curse of dimensionality. Increasing number of features in the clustering decreases the accuracy of clustering. Feature selection is necessary in many applications for effective information retrieval. The feature selection reduces the time consumption and memory wastage. The dataset may be imprecise, incomplete or uncertain. Rough sets deals with vagueness and uncertainty. Rough set theory (RST) has been successfully used as a selection tool to discover data dependencies and reduce the number of attributes contained in a dataset. Particle swarm optimization (PSO) is known to effectively solve large-scale nonlinear optimization problems. A semi-supervised hybrid feature selection based on PSO and RST for different datasets is proposed. Two feature selection algorithms namely PSO-quick reduct and PSO-relative reduct are applied for the different datasets. The simulation results of PSO-QR and PSO-RR show that hybridization of PSO with two rough set algorithms on semi-supervised data select features more effectively than rough set algorithms without hybridization of PSO.