Spectral clustering based on the sparse samples

Abstract

Clustering analysis is one of the key issues in the data mining technology. This paper proposes a method of spectral clustering based on the sparse samples (SCSS) that solves the problem of sparse sampling density. The algorithm firstly makes the data points into N times the original points in the l-nearest neighbors of each data point at random, increases the sampling density and selects optimal radius by many experiments. Finally, classifying the data points by spectral clustering based on local principal components analysis (PCA). Experimentation shows that SCSS produces accurate results in the case of no intersections and intersections and controls median misclustering rate for multi-manifold clustering.