A Bioinspired Proposal of Clustering Around Medoids with Variable Neighborhood Structures

Abstract

The artificial vision allows us to reduce a problem by means of techniques that have obeyed the study of the intelligence of living systems. A well-known technique is data mining and pattern recognition, which are disciplines dependent of artificial intelligence that from some data, allow the acquisition of knowledge and in particular, within data mining, a great application in the field of bioinformatics has been found. What is more, the big and diverse expansion of the amount of data produced by problems related to biological behavior has generated the necessity of constructing precise prediction and classification algorithms. The precision of classification algorithms can be affected by diverse factors, some of them considered generics in any automatic learning algorithm and, therefore, applicable to distinct research areas. These factors are the ones that have received attention in the field of automatic learning and pattern recognition, where different clustering algorithms are observed, in particular the automatic classification or better known as classification by partitions. In this scenery, is important to discover an analogy between the way that some living beings form groups to survive in their environment finding an optimal sequence or structure or grouping their objects or belongings, and a classification by partitions algorithm. The partitioning is an NP-hard problem, thus the incorporation of approximated methods is necessary. The heuristic that we expose here is Variable Neighborhood Search (VNS) focusing in the way that this heuristic does the search of neighbor conditions by means of neighborhoods to get a satisfactory solution, just like some living beings usually do it when they try to adapt to a neighborhood close to theirs or to the current space. In this work, we focus on describing in a bioinspired way, a technique of data mining known as partitional grouping with the inclusion of VNS with the purpose of finding approximated solutions for a clustering problem.