I mentioned in a comment above that I understand the concept of overfitting, but I asked this question due to a recommendation made to me. My model is correctly classifying new positive samples that were not used during training.

I understand the concept of sub-adjustment and the over-adjustment, I know that when performing a cross-validation, in the validation phase I get an accuracy of 94% for example and then in the test phase I get an accuracy of 100% so my model is not overfitted Because these values are not very different. However, one person told me to re-enter all the negative samples (not faces) of the training phase to the model that I have already trained and with that I could see if it is overtrained but I cannot know in what way I can conclude that from Negative samples

Thanks for your answer, your explanation helped me.

could you give me an example of stratified random sampling?