As Andrew mentioned after training the model, do error analysis and correct the label or improve the quality of the data and repeat the process. should the correction happen on training data?
Yes, you would start with training data. You would have to apply the same procedure of cleaning on the production data.
@satishnandi , thanks very much 