up sampling, down sampling, overfitting, underfitting

1 up sampling

increase the sample ratio

2 down sampling

decrease the sample ratio

3 overfitting

Recoganize

• Training accuracy >> Validation accuracy

• Performance varies wildly with small data changes

• Model memorizes training samples

Solution

• get more and better data
• simplify your model, reduce the complexity of your model
• regularization technique, L1/L2, dropout, early stopping, batch normalization,
• cross validation
• ensemble models

4 underfitting

 

Recoganize:

• Both training and validation/test performance are poor

• High bias, low variance

• Model makes oversimplified predictions

• Fails to capture clear trends in the data

Solution:

• increase model complexity
• add more features
• increase training epochs, iterations, 

• Reduce Regularization