Unsupervised Clustering Analysis Based On Multiple Optimization Algorithm

Clustering analysis is the process of automatically grouping data into a number of meaningful clusters according to a similarity metric without the premise of any prior knowledge. It is an effective tool for analyzing data, and some interrelations will be discovered through it, so clustering analysis has become one of the most popular techniques in data mining. Because clustering is also the process of making the objective function optimal constantly, thus the clustering problem can be transformed into an optimization problem. Inspired by this thought, many clustering algorithms based on Computational Intelligence with good optimization capability as well as high adaptive ability and robustness have been proposed, and their corresponding clustering models are built. A lot of references have verified that this kind of clustering methods can effectively solve the problem of data clustering, and can achieve good results. Some typical algorithms of Computational Intelligence, such as Genetic Algorithm, Immune Algorithms, Neural Network and Particle Swarm Optimization Algorithm and so on, have been applied to all kinds of clustering problems successfully.Under the context of the advantages of computing and storage brought by the rapid development of computer, a novel multi-group variant optimization algorithm-Multivariant Optimization Algorithm (MOA), is proposed recently. The main idea of MOA is to search the solution space comprehensively through global and local search alternately, and the global search is responsible for randomly exploring the whole solution space while the local search is responsible for exploiting a few potential local areas. In order to store and share the information gained during the search process effectively, a special structure is constructed and the search atom is positioned at the right position by following the operation rules of the table. In this paper, a new clustering algorithm based on MOA is proposed, and it is applied to data clustering analysis and its corresponding clustering model is built. We hope that we can find a new direction for clustering analysis problem and provide a valuable reference, and can put forward another more efficient and wider applicable clustering model by studying the proposed clustering algorithm, which is the research significance of this paper.This paper, at first, demonstrates the basic idea and the core framework of the MOA in detail, and introduces the multivariant structure, the alternate global and local optimization method of MOA, as well as the basic flow. Secondly, the clustering model of MOA is built based on MOA, and the way of encoding search atoms, the design of fitness function and the basic steps of this model are described in detail, then the clustering process of MOA is showed through two simple artificial datasets. At last, experimental simulation, the clustering algorithms based on MOA, k-means, GA and PSO are conducted on six standard datasets, and this paper draws a conclusion that the clustering algorithm based on MOA is a feasible method with good optimization ability and high stability in clustering problems.