I've recently made a binary image classification model with transfer learning. The model is used on an api and the predictions gets saved into a database. The problem is that it predicts images incorrect sometimes and these get saved into the db. My question is on how to label the incorrect labelled predictions correctly when i want to improve my model later on?

  • $\begingroup$ This might depend on the type of data/prediction, but I would use the same way you got the labels for the training data in the first place (e.g. a human looking at the data and assigning a label). $\endgroup$
    – Oxbowerce
    Apr 22 at 16:10
  • $\begingroup$ The data by itself comes from kaggle, where the data is split into 2 classes for binary classification. Right now a human looking at the data seems to be the only option, but this can't be done for in the future, since the company would expect around a few hundred rows of data a day. $\endgroup$
    – Enes Aygun
    Apr 23 at 19:19


Your Answer

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