Training and testing accuracy avoided overfitting and have 98% validation accuracy but classification report is always 0.01%?

Question

I have done fruit detection image classification problem using CNN i have done all the things upto training and fitting the model and my accuracy and validation accuracy are almost 100% but when i try to print classification report and confusion matrix from my model it always shows precision, recall and final accuracy is always 0.01% and confusion matrix is also bizzare. Why is this happenning please help me. Code is available at code section. Thank you.

This is my code for fruit classification

Thank you very much venkata krishnan it worked finally. Can you please tell me what does this shuffle. — Ganesh Kuikel, Jun 23 '20 at 07:35
i will post it as an answer, so that it helps future users. do accept the answer, if you find it useful. — venkata krishnan, Jun 23 '20 at 12:56

score 1 · Accepted Answer · answered Jun 23 '20 at 12:57

You test data is being shuffled, and that's why the classification report gives lower accuracy. Use

shuffle=False

for the test set while predicting, so that, you maintain the order of the prediction, in turn comparing it with the correct ground truth value.

Training and testing accuracy avoided overfitting and have 98% validation accuracy but classification report is always 0.01%?

1 Answers1