Research On Ant Colony Clustering Algorithm Based On Fuzzy Set

With the rapid development of information technology, there is no difficulty in obtaining and storing mass data any more. But if such mass data could not be analyzed and processed, they would be of no significance. So the present condition of "data drowning but starving for knowledge" urges the rapid development of data mining. Cluster analysis is an importance branch of data mining, and also the hot spot studied by the scholars at home and abroad. Since swarm intelligence occurred in the 1980s as a new and developing field, it has attracted the attention of researchers in several fields. Therein, the typical algorithm, i.e., ant colony clustering algorithm, will provide a good algorithm for cluster analysis as well.In this thesis, the conception, classification and method of cluster analysis are reviewed firstly, and then the related conception of fuzzy set and fuzzy clustering analysis are introduced. Next, the research status and typical algorithm of swarm intelligence are studied, including the primary idea, description and analysis of ant colony optimization, ant colony clustering algorithm and particle swarm optimization. Finally, the researches lay stress on ant colony clustering algorithm, and the author finds that there are some defects in basic model and LF algorithm, which would cause that dissimilar data object may not be picked up and similar data object may not be dropped, thereby the effect of clustering is influenced.In this thesis, aiming at the defect and regarding that "similarity" itself is a fuzzy concept, it is put forward to use the relative knowledge of fuzzy set theory to solve this defect. First of all, average distance isdefined;secondly the function of degree of membership of "similarity" is defined based on average distance;thirdly the pickup or drop of the data object is determined by the comparison between degree of membership and confidence level A;at last, the improved algorithm is realized by programming and its superiority is proved.Generally, the improved algorithm has the following advantages: (1) the number of parameters in LF algorithm has dropped and the meaning of new parameter A is easier to understand;(2) it is unnecessary to calculate the probability of pickup and drop in every iteration, so the amount of calculation is reduced and it is much closer to thinking process of intelligent creatures;(3) there is some reduction in the sensitivity of improved algorithm to similar parameter a .