A New Genetic-Based Hyper-Heuristic Algorithm for Clustering Problem

Abstract

Observations over recent studies indicate that most of the methods and algorithms used to deal with clustering problems are based on hybrid metaheuristic and metaheuristic algorithms to rectify the solutions. However, these approaches are restricted by the number of heuristics. Hyperheuristic algorithms are new generation of metaheuristic algorithm that use a collection of low-level search strategies and high-level heuristics which works in heuristic space and solution spaces, while metaheuristic algorithms just work in solution space to find better solutions. The main goal of this research is to propose a hyperheuristic framework for clustering problem which is able to optimize the difference of all data objects of one cluster from their respective cluster centres. The proposed hyperheuristic clustering algorithm has used from pool of meta-heuristic and heuristic approaches and mapping between solution space and heuristic spaces is one of the prevalent and powerful methods in the optimization domains. By mapping the solution spaces into heuristic space, it would be possible to make easy decision to settle data clustering problems. Our suggested hyperheuristic clustering works into three major spaces including high-level space, low-level space and problem space. The experiments of this study have proven that the suggested method has successfully generated efficient and robust clustering results.