2
$\begingroup$

I've a dataset with about 123 images (two categories, 19 defect and 104 no defect).

I've to implement a classifier so I've decided to split my data in train (70% of all data), validation (20% of all data) and test (10% of all data) set.

At the end I've:

  • train: 72 no defet + 13 defect
  • validation: 20 no defect + 3 defect
  • test: 12 no defect + 3 defect

For the training set I've decided to perform data augmentation in order to increase the number of samples. Having validation and test set so much small with few samples for defect could be a problem? And if yes could I improve this situation in some way?

Improve this question
$\endgroup$

2 Answers 2

Reset to default
4
$\begingroup$

Having validation and test set so much small with few samples for defect could be a problem?

It, typically, is a problem. With such small numbers, performance evaluations may have large variances. Cross-validation in both hyperparameter/model selection and test evaluation is advised for more robustness (i.e. nested cv).

Improve this answer
$\endgroup$
0
$\begingroup$

If the quantity is equal to the tests then it would be true however if both waiver and quantity the quality would be built-in and no test sets would be required. Professor Bill Maher

Improve this answer
$\endgroup$
1
  • $\begingroup$ Can you edit to explain that as at the moment I find it hard to see what you are driving at. You could also say how you think it adds to the existing answer. $\endgroup$
    – mdewey
    Commented Jun 19, 2022 at 14:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.