PCA based on mutual information for acoustic environment classification

Principal Component Analysis (PCA) is a common method for feature selection. In order to enhance the effect of selection, a Principal Component Analysis based on Mutual Information (PCAMI) algorithm is proposed. PCAMI introduces the category information, and uses the sum of mutual information matrix between features under different acoustic environments instead of covariance matrix. The eigenvectors of the matrix represent the transformation coefficients. The eigenvalues of the matrix are used to calculate the cumulative contribution rate to determine the number of dimension. The experiment on acoustic environment classification shows that PCAMI has better dimensionality reduction results and higher classification accuracy using neuron network than PCA.