Research On Application Of Multi - Core Function FCM Algorithm In Data Mining Clustering

Data Mining has been recently the topic view of Artificial Intelligence and Database studying as a tool of extracting implied information, discovering complicated relations and evaluating value. In the face of challenges of extracting and discovering the knowledge and rules which are used to support business decisions and future planning and guide the enterprise developing and science research from the mass and complex data, Data Mining plays an important role in the various walks of life. Clustering algorithm is the core technology which categories a synthetic data set into individual clusters artificially to reveal the true distribution of data set as a statistics method and the important preprocessing stage and duty of Data Mining and helps people deal with data. However, the fuzzy clustering algorithm has been a rapid development in Data Mining which is used broadly in pattern recognition, decision analysis and data clustering as one of the most significant clustering algorithms. In this paper, the FCM clustering algorithm and KFCM algorithm are learned and analyzed thoroughly and the Fuzzy C-means clustering based on Multiple Kernels-MKFCM is proposed upon the drawback of KFCM algorithm.In order to compensate for the shortage that KFCM algorithm using a single fixed kernel is not able to categorize the data set including categories unbalanced sizes and densities, MKFCM algorithm utilizes a more flexible and suitable way which constructs a new Kernel from a number of Gaussian Kernels and learns a resolution specific weight for each Kernel function in each cluster to replace the single kernel and maps data point to a high dimensional feature space through an optimal convex combination of kernels and then completes the computation in an low dimensional space with kernel trick completely under an unsupervised way. The resolution specific weight is the key factor influencing the creation of cluster which reflect the density fitting relation between kernel and cluster. A low value indicates the kernel is not relevant for the density fitting of cluster and has no significant impact on the creation of this cluster; Similarly, a high value indicates that the kernel is highly relevant for fitting the points in the cluster. The fixed set of kernels with different resolutions can cover the spectrum of entire data and take into account the variations of the different clusters which is able to divide the data set into several subsets very well. Above all, MKFCM algorithm has an efficient and better application.Simulation results show that KFCM algorithm using the fixed single kernel or the kernel constructed as the average of several Gaussian kernels is not able to categorize the set. Compared with KFCM algorithm, MKFCM algorithm using a set of kernels with different resolutions can categorize the data set including categories with unbalanced sizes and densities correctly; In the end, MKFCM algorithm is applied in the package customization of colleges and simulation result indicates that MKFCM algorithm can categorize the three clusters correctly which prove the suitability in real life of network big data mining and has a significant future in Data Mining area.