Research And Implementation Of Clustering Algorithm Based On Multi-manifold

Manifold learning plays a vital role in data analysis and artificial intelligence techniques.Traditional manifold learning algorithms are based on high-dimensional mono manifold models,however,due to the existence of multidimensional multi-manifold models and the complex relationships between them,the performance of manifold learning algorithms is affected to a certain extent,resulting in a decline in performance.In order to effectively process high-dimensional multi manifold data,we must adopt multi manifold clustering techniques,especially for multi manifolds with multiple intersecting points.(1)In this paper,a label propagation algorithm LPAMMC based on the manifold hypothesis is proposed.The traditional label propagation algorithm constructs a connection matrix through the similarity matrix to achieve the clustering of well-separated data,but it cannot effectively cluster the overlapping multi manifold data.In order to solve this problem,this paper uses the local principal component analysis algorithm to determine the intersection region of the intersecting multi manifold data,and based on the mixed probability principal component analysis(MPPCA)model and the topology of the multi manifold,preliminarily divides the overlapping submanifolds,constructs the "must-link" and "cannot-link" clustering constraints,and then uses the constraints to construct a propagation matrix suitable for intersecting multi manifold data to implement the label propagation algorithm.(2)In this paper,a density core normal vector algorithm NVLDK is proposed.At present,most density clustering algorithms can only process clustered data,and are not suitable for manifold data.To solve this problem,a normal vector algorithm NVLDK based on density cores is proposed.First,to introduce the concept of natural neighbors,the neighborhood information of each point is calculated.The areas where the manifold intersects are then identified using the natural neighbor density.For non-intersecting regions,local density cores are used to represent discrete points to form a pile of data.Finally,the normal vector information of each heap is combined and separated,and the remaining discrete points in the intersection region are clustered using Euclidean distance.(3)This paper introduces a multi manifold clustering system based on MATLAB,which consists of five modules: data import module,local PCA recognition,MPPCA blocking,LPAMMC algorithm,and parameter modulation module.Provide researchers with an easy-to-use tool to conduct their research.