This does attribute-value learning once the features have been selected. Almost any conventional learner can be used. For example, we may apply a rule-learning system to the dataset, a decision tree learner or any other attribute-value learner. We applied c4.5rules, a simple rule-learning system [Qui93], to the Blues and Reds domain. It came up with the rules shown in figure 5.6.
Figure 5.6: Rules for telling blues and reds apart.
As can be seen, the rule is very simple. In general, however, this will not always be the case.