Validation and formatting phase
Knowledge extraction from data (KDE)

The models extracted, particularly by supervised learning methods, cannot be reliably used directly. We need to evaluate them, i.e. put them to the test of reality and assess their accuracy. The usual procedure is to estimate the model's error rate as accurately as possible. In this way, the user can decide whether or not to apply the prediction model, in full awareness of the risks involved. The error rate is generally calculated from the confusion matrix (table 1 ). This gives the percentage of assignment to the different classes according to the original classes.