Label

This Label option is only available when Classification is selected as the problem type:

• If the target column has only two unique values, this means a binary classification, hence mere existence of two classes (such as the case in the above image). If there are more than two classes in the target column, this is a multi-class classification.

• TAZI will label the class with most frequent values as normal and the rest of the classes as anomaly.

• TAZI will assign class weights inversely propotional to the value counts in class type. This will make classes with few instances maintain their importance during training of the model.

• You can assign class weights manually or accept the class weights recommendation given by Profiler.

Let’s see how the LABEL option looks exactly when the problem type is of binary classification: