CODEWITHSUNDEEP
pythontensorflowkerasevaluationconv-neural-network
baddy
baddy

Reputation: 417

Could not understand if my CNN model for multiclass classification is overfitted or not?

Good training, testing and validation accuracies but strange historical accuracies behavior for model: Here is the summary of my model :

enter image description here

I performed the execution and prediction tasks and I've got the next confusion matrix :

enter image description here

while the Accuracy behavior was the next :

enter image description here

I can not understand if this is overfitting or underfitting or a normal behavior?

Adding the loss plot to clarify more in the next

Figure

Thank you in advance for any useful information and help !

Upvotes: 1

Views: 241

Answers (2)

Sebastian Baum
Sebastian Baum

Reputation: 365

As you already mentioned your training is doing well.

First of all I recommend you to check a prediction by yourself with test-data. The Validation-loss will converge until a specific value. It may looks a little bit variance but you need as reference the y-Axis. The ups and downs of the last epochs are between 91% and 94% which is not really much in reference to 100% (maybe change the y-Axis).

Upvotes: 0

Gerry P
Gerry P

Reputation: 8092

Does not look like over fitting. Your training accuracy is increasing and so is the AVERAGE test accuracy. Over fitting is when the test loss improves, then plateaus and then starts to increase. It is best to look at loss metrics to monitor this. It is typical that once the training accuracy gets high the test loss will oscillate to a small degree. You can test for over fitting by varying the drop out rate and see the effect on test loss.

Upvotes: 1

Related Questions