is a way of constructing clusters of objects whose properties are described by binary data variates (that is, variates that take one of two values, usually designated by zero and one). The class predictor is defined to be a variate that contains an entry for each data variate, storing the value that is more frequent in the class for that variate.
The method aims to maximize the sum over the classes of the number of agreements between units of each class and their class predictor. When several different classifications give the same maximum value, the method then minimizes the sum of the number of correct predictions for each unit when predicted by any of the class predictors of the classes other than the one to which the unit is assigned. (See