-1

I'm trying to train CNN model for MRI Classification. As you can see the train loss is smaller than validation loss. My question is :IS training loss >> validation loss that we can say we have Overfitting?

enter image description here

desertnaut
  • 57,590
  • 26
  • 140
  • 166
Dhia Mansour Khouja
  • 1
  • 1
    I’m voting to close this question because it is not about programming as defined in the [help] but about ML theory and/or methodology. – desertnaut Jun 19 '21 at 12:21

1 Answers1

1

If the gap between them is quite big, then your answer is yes. But from your image description, your loss from train and validation doesn't so big, and your accuracy on train and validation is acceptable. So your model isn't suffering from overfitting.

Trong Van
  • 374
  • 1
  • 13
  • thank's..that's very comforting :)) .. one other question PLZ : when I tried to classify images of the Test Set I got this: The value of the loss function on the test data set is: 0.8562 The categorical accuracy of the predictions on the test data set is: 0.7396 The categorical AUC (i.e., average curve across classes) of the predictions on the test data set is: 0.9068 Do u think that is ok ! – Dhia Mansour Khouja Jun 19 '21 at 04:55
  • You need to compare the accuracy between train and test. If your train accuracy is higher than the test set (the gap between them is big, how big is determine by yourself, e.g., 5%), it is overfitting. More details [here](https://datascience.foundation/sciencewhitepaper/underfitting-and-overfitting-in-machine-learning), section `Detecting Overfitting or Underfitting`. Hope it helpful to you. – Trong Van Jun 19 '21 at 05:15
  • Merci beaucoup :)) – Dhia Mansour Khouja Jun 19 '21 at 05:36
  • You can upvote the answer if it is helpful. Glad I could help. – Trong Van Jun 19 '21 at 05:45