Feature Selection and Clustering By Multi-Objective Optimization

In this paper, feature selection and clustering is formulated simultaneously by using evolutional multi-objective algorithm. Archived multi-objective NSGA-II is hybridized with k-medoids algorithm to use global searching capabilities of GA with local searching capabilities of k-medoids for suitable centers of clusters and selecting suitable subset of features identifying the correct partitioning. Number of clusters should be determined as an input parameter by user. After determining number of clusters, archive string be generate randomly. In every solution of archived, center of clusters and features is determined. Objective functions are inter-cluster distance, intra-cluster distance and number of feature selection. Three objective functions are optimized simultaneously for partitioning and feature selection. Crossover and mutation operators are modified to solve the problem. In order to selecting final solution from pare to front, are modified to solve the problem is calculated. The Proposed algorithm were compared with other three clustering algorithms on seven UCI standard datasets and could improve results averagely 0.09 percent compared to FeaClusMoo, 0.28 percent compared to VGAPS-Clustering and 0.49 percent compared to K-means.