The training cases with the selected attribute value known are split in the child nodes. The training cases with
exactly that attribute value missing will appear in every child node. Each of these training cases in each child node
will have two associated parameters: confidence and error (real numbers). Obviously, training cases with that value
known have confidence one and error zero associated with their class. The process to obtain the values of these
parameters is performed within a Decision Theory enviroment [33]. The attribute with missing values is interpreted
as a statistical parameter. We also take into account the class to which the training case with an unknown value
belongs. With this information, we calculate the Bayes’ probability (confidence parameter) and the Bayes’ risk (error
parameter). We calculate the Bayes’ risk through the loss matrix obtained from the distance between the different
values of the attribute.