0

I have been trying to implement DenseNet on small dataset using k-fold cross validation. Training accuracy is 94% ,validation accuracy is 73% whereas test accuracy is 90%.I have taken 10% of my total dataset as test set. I know some overfitting is present, but how can validation accuracy be greater than test accuracy?

srij
  • 13
  • 4
  • 1
    Can edit your post to include the total number of samples, the number of samples in your training and test sets, the value of k in k-fold, and finally the standard deviations of your training and validation accuracy? – Snehal Patel Nov 26 '22 at 15:56
  • 3
    Also, your title and question imply `val > test` but your numbers indicates `val < test`. Please verify that your numbers are correct and clarify your question. – Snehal Patel Nov 26 '22 at 16:02

0 Answers0