Improved Particle Swarm Optimization Projection Pursuit Clustering

With the rapid development of information technology, the quantity of information handled by the users is increasing at the different speed. Facing such a large database of information, the how to gain knowledge need by people is a problem which should be solved urgently. The clustering analysis, as a kind of data analysis tool appeared in this case. So-called clustering is a process, which divides a physical or abstract data into some subsets making up similar objects, its purpose is to make the same category of sample belongs to the similarity between different categories of higher, and the large difference between samples, so as to further analyze the data.In recent years, with the application of clustering, high dimensional data clustering is becoming more common, and more important. The traditional clustering algorithm is facing challenges, when the dimensional of data clustering is high:with the dimension increased, the calculated quantity rapidly does; and there are many "dimension disaster" of sparse midpoint in high dimension space; The robustness of the clustering method has lowed, when the dimension enlarged. The classic clustering algorithm, such as K-Means and K-Medoid, has not performance well, when it directly deal with high dimensional data. So researchers has paid attention to the dimension-reduction method, such as principal component analysis algorithm, but this method easily loss the original information of data. As the continuous development of research direction, the projection clustering method was proposed.Projection clustering projects the data into the low-dimensional subspace by mapping projection, then clusters in the subspace using various methods, which can effectively reduce the dimension of data sets, reduce the complexity of the data processing, then also can research and analysis the high dimensional data. The performance is good and robust, and its result is accuracy.The thesis has referenced many interrelated document and knew the principle and application of the clustering algorithm, some improved algorithm and application are researched, as follows:Firstly, combining the mentioned several kind of intelligent algorithm to the traditional k-means clustering method, the individual encoding, the structure of fitness function and the calculation method are provided. Then through the simulation experiment, according to the analysis of experimental results of different intelligent optimization algorithm in clustering problem, find out the performance is suitable for solving the problem, a kind of clustering algorithm.Secondly, this thesis used intelligent algorithm to optimization projection pursuit clustering model. Expounded the projection pursuit of background and development, and its basic ideas, characteristics and the mathematical model, this paper introduces intelligent optimization algorithm and the projection pursuit of combining theory and applied in the cluster analysis. The algorithm in detail, and the realization process simulation, test results and basic will k-means clustering results demonstrated that this new mix of the cluster analysis algorithm is effective, and shows that the quantum particle swarm optimization algorithm can make projection pursuit clustering model of optimization.Finally, the basic projection pursuit clustering model is improved, in particularly the improvement of the objective function. And the related experiments verify the feasibility of the algorithm. Through the comparison with the basic model, the improvement strategies are proved feasible, and the operation is efficient and clustering results have improved, and the experiment result proves that the quantum particle swarm optimization is the optimal algorithm for projection.In this paper, on the basis of the above work to find an effective clustering algorithm, which is based on the quantum particle swarm and improvement projection pursuit clustering algorithm, many simulation experiments prove this algorithm is effective and feasible. By using this algorithm and the data of bioinformatics cluster analysis, such as breast cancer cells, Iyer gene expression, the result is still ideal. Thereafter, the clustering algorithms will be applied in more fields of biology.