On Selection of Features in Pattern Recognition

For the case of complete a priori information, a procedure is found for selecting a combination of independent features in two formulations. In the first formulation, the minimum number of features ensures a specified risk level in pattern recognition; in the second formulation, the average risk is minimized by selecting an array of features, with allowance for the cost of obtaining and of processing the features. An example of feature selection is given for the case of two-alternative recognition.