If we applied a confused matrix for the CNN models. Tricky results.

[mayadiCS]

The section is really interesting. That being said I’m a little worried that we are only judging our model performance based on accuracy + checking Ioss and accuracy curves. So I attempted creating a confusion matrix for model_7 and the results are strong only for the True Positive (TP) class and not the True Negative (TN) (i.e. in the diagonal the 2nd entry is 0). We are only good are classified one class. I am genuinely curious if I am worrying for no reason? (I.e. is this common when building Conv based models. Since our output Dense layer is 1) or is there an issue and our model is actually only say learning features for 1 class only and when an images doesn’t fit that it guesses “not my class, hence the other”?
(see attached screenshot)

PS: I could be very wrong in the way I coded the confusion matrix. If that’s the case. I definitely welcome and would appreciate that feedback.