In classification, suppose you have 1 image labeled as cancer and 99 labeled as not cancer, you can just divide the loss weight of "not cancer" by 99. Then you can train the model as this will help fix the label imbalance problems somewhat.

But in regression, it's not trivial what number I need to use to divide. Suppose you regress the probability of cancer from 0 to 1 instead of classifying it, and someone has collected the dataset with probabilities. The dataset has cancer around 0.1 about 99% of the time and you have cancer around 0.9 about 1% of the time. But the dataset also have labels like 0.26, 0.3634, 0.753, 0.5, etc. It's not clear what loss weight you need to use for cancer=0.26 or any other cancer probability.

Is there any formula to figure out loss weight for regression problems that have imbalanced labels?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.