Clustering Analysis Methods Based On Swarm Intelligent Algorithm

To overcome the faultiness of the available clustering algorithms for applications in data mining,by combining with the swarm intelligent algorithms, the traditional cluster analysis algorithms are systematically improved and innovated in this thesis. And the new clustering analysis methods based on swarm intelligent algorithms are proposed, then analysis the performance of the methods. The experimental results illustrate the effectiveness and good Performance of the proposed new ideas and new methods on cluster analysis. In sum,the main researches in this thesis are given as follows:1) Clustering analysis method based on population migration algorithm was proposed. Firstly put the objects in two-dimensional surface stochastically, each object has a stochastic initial point; each object can move in the surface, and then survey the local environment community similarity of the object. Transformed the community similarity into the income/attraction function through the transfer function, and then realized the organization clustering process according to the income/attraction function.2) A new mix clustering algorithm based on artificial fish school algorithm was proposed. Artificial fish school algorithm doesn't need the prior knowledge; it uses stochastic traversal principle for clustering analysis. K-means algorithm needs initial division, uses determined/heuristic principle for clustering analysis. Firstly uses the artificial fish school algorithm for each data object; then inspects the cluster results, selects the results as entrance point of the K-means algorithm; finally uses K-means algorithm to clustering analyses.3) A dynamic fuzzy clustering method based on artificial fish school algorithm is presented. By importing a fuzzy equivalence matrix to express the dissimilar degree between any two data, and mapping the high dimensional samples to two dimensional planes. And then using artificial fish school algorithm to optimize the coordinate values, making the Euclidean distance of the samples approximate to the fuzzy equivalence matrix gradually, finally realized the fuzzy clustering.4) A spatial clustering analysis method based on differential evolution algorithm was proposed. Uses mutation, crossover and selecting operator of differential evolution, and combine with the spatial data unique characteristics, then the new individuals are produced unceasingly from the processes, guarantee the multiplicity. New individuals who produce from the evolution, namely the individuals from the variation and crossing operators insert K-means algorithm, thus might speed up the convergence rate. Simultaneously increase some flexible changes in operation; the experimental results proved the effect was good.