Multiclass parametric decision-making processor for classification of patterns with missing descriptors

In the paper, the problem of classifying pattern vectors with missing descriptors using the parametric minimum-error decision rule for normally distributed classes is considered. A computationally efficient method for determining the optimal parameters of the classifier for operating in any subspace of the pattern space is proposed. In general, for any number of missing descriptors satisfying q < n/2, where n is the dimensionality of the complete pattern space, the method affords considerable saving in both computer time and storage requirements. Consequently, the cost of implementation of the classifier is substantially reduced.