Research On Fuzzy Clustering Algorithm Based On Granular Analysis Principle

As an unsupervised learning method, clustering analysis is an important research orientation in intelligent computing domain. Meanwhile, clustering technology is also an important analysis tool and method of data processing in data mining. With the development of modern information technology rapidly, clustering analysis has become research hotspot in machine learning, pattern recognition, data mining, intelligent computing, information retrieval, and so on.The design of clustering model and clustering algorithm is the key step during the whole clustering analysis. By designing the different clustering models, we can obtain the different clustering algorithms. At present, the main clustering algorithms include partition, hierarchy, density, grid, model and so on. Clustering analysis implies the granularity idea; therefore, the research of combining clustering algorithms with granular analysis principle causes extensive attention at home and abroad.This thesis introduces the basic concept and fundamental knowledge of clustering analysis and information granularity, generalizes, analyzes and researches the typical algorithms and fundamental idea of clustering analysis. Combining coupling degree and closing degree of information granule, this thesis introduces granular analysis principle into clustering algorithm, and makes advanced research on fuzzy clustering algorithm. The main content contains some aspects as following:(1) Based on the theory of minimal square error, this thesis presents a measurement of coupling degree and closing degree of information granule, makes a valid evaluation on information granularity from different granular space by computing coupling degree and closing degree.(2) Given the largest number of initial clusters, the centre of initial clusters is constructed through max- min distance algorithm based on the certain number of clusters. Next, by introducing the membership of data sets belonging to clustering center, the very thesis clusters on data sets. Finally, by using the measurement of coupling degree and closing degree of information granule, the thesis obtains better clustering results through fuzzy iteration method, and meantime finds the optimal clustering number.(3) With IRIS standard test data sets and simulation data sets respectively measuring the algorithm, the results showed that the proposed method reflected better clustering affection than the traditional max-min distance algorithms and fuzzy C-means algorithm, thus proved the validity of algorithm proposed in the thesis.This thesis combines clustering algorithms with granular analysis principle and evaluates clustering results by using the measurement of coupling degree and closing degree of information granule to acquire ideal results. Besides, this thesis not only promotes theory research of fuzzy clustering analysis but also obtains some results having important applied value.