Timeline for How to fix class imbalance in training sample?
Current License: CC BY-SA 3.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 16, 2020 at 11:08 | history | edited | CommunityBot |
Commonmark migration
|
|
| Feb 28, 2018 at 10:18 | comment | added | Kasra Manshaei | Glad it helped :) Actually resampling (rebalacing) is the way to keep those errors fairly small. You kind of saw it from the other way around. If your classes are balanced then you get a better Precision and Recall. And about distribution: Usually the assumption in machine learning is that train and test are from similar distributions so do not spend much time on that. The situation in which they are not similar is a branch in Machine Learning called "Domain Adaptation" or "Transfer Learning" | |
| Feb 27, 2018 at 18:49 | comment | added | Learning is a mess |
Many thanks for the detailed answer. So, if I get it right, if the cost of misclassification is the same cost(false positive) = cost(false negative), then I can use the accuracy as metric and rebalancing should only be done to match the distribution of the test sample. Is that right?
|
|
| Feb 27, 2018 at 18:39 | vote | accept | Learning is a mess | ||
| Feb 27, 2018 at 17:17 | history | edited | Kasra Manshaei | CC BY-SA 3.0 |
typo
|
| Feb 27, 2018 at 17:11 | history | answered | Kasra Manshaei | CC BY-SA 3.0 |