The Hybrid Clustering Algorithm Based On Nuclear Research

Fuzzy clustering algorithm is an unsupervised clustering algorithm. For example, thetypical FCM and PCM clustering algorithm have a wide range of applications in real life.However, the basic fuzzy clustering algorithms have their own shortcomings:FCM is sensitive to noise, PCM coincides at the center of the classes. In view of this case,hybrid clustering algorithm can be put forward. This algorithm retains their individualadvantages but avoid their disadvantages. The ideal result is obtained through experiments.Because of the complexity of the real data, for a lot of high-dimensional data andheterogeneous data sets, the hybrid clustering algorithm shows its disadvantages. Based onthe above reasons, Mercer kernel is introduced in this paper, a clustering algorithm of thehybrid model based on kernel is proposed, Through the simulation experiment, the validity ofthese algorithms is confirmed.In this paper, specific content arrangement can be divided into the following3points:1. In view of the preliminary research on the kernel theory, in the first and the second chapterof the paper, we introduces the research background of kernel method and its principle.Furthermore, we summarize the clustering analysis of the theoretical basis and methods. Lastly，KPCM1and KPCM2kernel clustering algorithms are present ed.2. The hybrid model of FCM and PCM can overcome the shortcomings while they areclustered separately. It has great improvement in the clustering effect. But the effect is not sowell for the samples without obvious characteristics. In order to overcome thesedisadvantages, in the third chapter, we introduce the Mercer kernel and propose a new modelcalled Kernel-based hybrid c-means clustering (KIPCM). This model makes it possible tocluster the data which is non-separable in the original space into homogeneous groups in thekernel space by using kernel function. We get better center values and higher correctclassification rate through numerical experiment.This confirms the feasibility and validity ofthe algorithm in this paper. 3. For multiple data sources or heterogeneous data sets, single kernel function has its ownadvantages in processing, but has its own shortcomings at the same time. For example, theinput space is the space of two vectors: the first vector obeying polynomial distribution, whilethe second vector obeying gaussian distribution. In this case, using a kernel function to clusteralone is insufficient. So they need be grouped up to form multiple kernel function forclustering analysis. In the fourth chapter, we discuss some combination methods of the singlekernel functions. Next, the multiple kernel function is introduced in the clustering, expectingto obtain a better clustering effect.