On k-medoid clustering of large data sets with the aid of a genetic algorithm: background, feasiblity and comparison

Abstract

A novel approach to the problem of k-medoid clustering of large data sets is presented, using a genetic algorithm. Genetic algorithms comprise a family of optimization methods based loosely upon principles of natural evolution. They have proven to be especially suited to tackle complex, large-scale optimization problems efficiently, including a rapidly growing variety of problems of practical utility. Our pilot study lays emphasis on the feasiblity of GCA — our genetic algorithm for k-medoid clustering of large datasets — and provides some background information to elucidate differences with traditional approaches. The experimental part of this study is done on the basis of artificial data sets and includes a comparison with CLARA — another approach to k-medoid clustering of large data sets, introduced recently. Results indicate that GCA accomplishes a better sampling of the combinatorial search space.