New Covariance-Based Feature Extraction Methods for Classification and Prediction of High-Dimensional Data