A method for principal components selection based on stochastic matrix

Principal component analysis (PCA) is a feature extraction method of mapping the original characteristic term space with high-dimensional sparsity into the low-dimensional principal component space, by utilizing the principle of variance maximization. In random matrix theory, principal components are treated as random variables, and data samples are the observations of random vectors. When there are a large amount of data samples, the observed values of principal components obey the normal distribution. We proposed a method of principal components selection based on the differential distribution of principal components in different classes. Experiment on text classification indicated that the method of principal components selection based on random matrix theory can effectively determine the principal components. This not only improves the classification performance, but also reduces the computational cost.