A classification model on the basis of partial information about features in the form of their mean values

A classification model under partial information in the form of expectations of features is proposed. It is based on the minimax and minimin decision strategies. The discriminant function is computed by maximizing (minimizing) the risk functional as a measure of classification error over a set of probability distributions with bounds determined by the information about features, and its minimizing over a set of parameters. An algorithm is reduced to the parametric linear programming.