Instance Selection Using Multi-objective CHC Evolutionary Algorithm

Abstract

Data reduction has always been an important field of research to enhance the performance of data mining algorithms. Instance selection, a data reduction technique, relates to selecting a subset of informative and non-redundant examples from data. This paper deals with the problem of instance selection in a multi-objective perspective and, hence, proposes a multi-objective cross-generational elitist selection, heterogeneous recombination, and cataclysmic mutation (CHC) for discovering a set of Pareto-optimal solutions. The suggested MOCHC algorithm integrates the concept of non-dominating sorting with CHC. The algorithm has been employed to eight datasets available from UCI machine learning repository. The MOCHC has been successful in finding a range of multiple optimal solutions instead of yielding a single solution. These solutions provide a user with several choices of reduced datasets. Further, the solutions may be combined into a single instance subset by exploiting the promising characteristics across the potentially good solutions based on some user-defined criteria.