Cluster ensemble selection based on relative validity indexes

Abstract—There are variety of methods available to solve multi-objective optimization problems, very few utilizes criterion linkage between data objects in the searching phase, to improve final result. This article proposes an evolutionary clustering algorithm for multi-objective optimization. This paper aims to identify more relevant features based on criterion knowledge from the given data sets and also adopts neighborhood learning to improve the diversity and efficacy of the algorithm. This research is an extension of the previous work named neighborhood learning using k-means genetic algorithm (FS-NLMOGA) which maximizes the compactness of the cluster and accuracy of the solution through constrained feature selection. < Final Year Projects > The proposed objective finds the closest feature subset from the selected features of the data sets that also minimizes the cost while maintain the quality of the solution. The resultant cluster were analyzed and validated using cluster validity indexes. The proposed algorithm is tested with several UCI real-life data sets. The experimental results substantiates that the algorithm is efficient and robust.