Feature selection for classifying high-dimensional numerical data