my curve of validation and training accuracy is something like this .Validation accuracy is decreasing after a certain point but it again increased with much higher values .
https://drive.google.com/file/d/1ns6uh673dbajX_Tku8VTLjPJGIv7tQ2h/view?usp=sharing
At what values of epoch should I stop
Accuracy is not always the best metric to work on , but if you are using it , then take the epochs values that has highest validation accuracy and confirm that it is comparable to training accuracy also.
things to change :
- yabel to accuracy
- Add regularization to your model , and train for more epochs , say 50-100 ,and then see what is the optimal solution.
- You can also use Callbacks like EarlyStopping to stop when you metric is not improving for given iterations.
other than accuracy what metrics are there and which metrics is used in which situations ?
Accuracy is generally used in cases where the class distribution is balanced and you have small data.
where as other like , f1 score , precision , recall and etc… are the ones that actually help in understanding the model performance , both on balanced and unbalanced datasets.
I hope this helps.