Research On Fuzzy C-Means Clustering Algorithm

Clustering analysis is an unsupervised method of data mining. It can explore the potential and valuable information by using the suitable algorithm, and improve the quality of data analysis, as well as provide a scientific way for the other data analysis and reprocessing data. And in the real world, the boundary of lots of objects is ambiguous, so as to classify the objects, fuzzy clustering analysis is generated. Fuzzy C-Means clustering analysis is an important branch of fuzzy pattern recognition.Firstly, fuzzy clustering algorithm for interval data was studied. The steps of the Fuzzy C-Means algorithm that could control the impact to the length of the interval data were presented. But the algorithm cannot find the different shapes and sizes, and the effect is not accurate, so an adaptive distance to solve the problem is presented, which can deal with the irregular fuzzy subset and prevent outliers from affecting the result of the algorithm in consideration of the attribute of the objects and the length of the interval data; At the same time the IFCM algorithm makes the clustering results more accurate.Secondly, because Fuzzy C-Means clustering algorithm could not deal with the updated data directly, so an Incremental Clustering Algorithm Based on Adaptive FCM is presented. When inserting or deleting an object in database or data warehouse, the prototype and the number of the subsets will be altered, FCM cannot cope with that in real time. In order to solve the problem, the paper presents an Update-center algorithm and Split cluster algorithm. Also, both of them excel in the real-time tasks.Finally, AIFCM algorithm uses set theory, the density, cohesion, separation, Update-center algorithm and Split algorithm to deal with the incremental data which FCM cannot resolve except for clustering all the data from original to the incremental data; In the process of clustering, AIFCM can alter the final number of the clusters dynamically and find the irregular shaped fuzzy sub cluster when merging the similar subsets, so AIFCM can both find the spherical and irregular shaped subsets. The AIFCM algorithm can cope with the renewed data and filter the outliers.