Study Of Elliptic Ring-shaped Clastering Based On Adaptive Genetic Algorithm

Clustering analysis, which is one type of multi-variable statisic analysis and an significant branch of unsupervised pattern recognition, is defined as the study and processing of the characterization of a given set of data with mathematic approach. According to certain criterion, this technique partitions a sample cluster without group marked into some groups, so that similar samples are assigned to one group as possible as much, while dissimilar to different ones.Genetic algorithm (GA) derived from the computer simulated study of biosystem, which is one kind of contemporary optimization techniques, and features functions such as the overall and effective parallel optimization, high robustness and generality, and no demand for specific information of problems, or else. Disciplines like mathmetics, physics, biology and computer science and etc. are all embodied in the algorithm, which sheds new light on the solution of complicated problems.In the present study, the genetic algorithm is creatively employed to the optimization of clustering objective function. The application of Adaptive Genetic Algorithm(AGA) to the detection of elliptic ring-shaped cluster successfully presents convergence into local minima, which might result from the traditional simple clustering method. The study result proves that the newly constructed algorithm is very robust.The dissertation is divided into four chapters, with contents as follows: Chapter 1 is the introduction, which introduces the fundamental issues of cluster analysis and the necessity and rationality of the application of GA to clustering objective function optimization as well.Chapter 2 looks back on the developmental course of clustering algorithm, which mainly centers on the introduction and derivation of the Fuzzy C-means Algorithm(FCM) and disscusses respectively the strong and weak points of the traditional clustering algrorithms. In the end, the FCM algorithm is employed to process some data of funds, just in order to test the validity of the algorithm.Chapter 3 focuses on analysis of the defects of the Fuzzy C-ellipsoidal shells algorithm(FCE) and the cause for these defects, that is, the FCE algorithm relies excessivelly on the FCM algorithm for the definition of the initial value and the iterative method rooted alternating optimizing strategy, which inevitably leads to the dissatisifactory partition results for the FCE algorithm in use. Therefore, we creatively combine the FCE and GA algorithms, and propose the Adaptive Genetic algorithm---Fuzzy C-ellipsoidal shell algorithm (AGA-FCE).The new algorithm is tested on several examples, which show that the AFCE algorithm proposed to be robust in the presence of noise in the data. The algorithm also prevents the weakness of local minima caused by the traditional algorithm in the optimization scheme. The testng results show the strong effictiveness of the algorithm.The last chapter concludes the whole dissertation. The author concisely lists the controbution of the present research and suggests the following research prospective.