Isn't this what the off-diagonal elements of the confusion matrix from model validation tell you? If two classes A and B are overlapping from the perspective of the classification technique you are using then you will see a high rate of misclassification of A as B and vice versa.
If that's so then you could decide to filter one of those classes out of the training set, but I'm not sure in what circumstances that would be valid: the training set should be representative of what you will see in the test set, surely.
----- Original Message -----
Date: Tue, 23 May 2017 12:57:54 -0400
From: Michael Calve <[hidden email]>
To: [hidden email] Subject: [Wekalist] Too much overlapping data between multiple classes
Content-Type: text/plain; charset="utf-8"
Is there a way/algorithm/filter that compares training data before including it as part of the model, in order to make sure that classes are not overlapping or too similar.