ML internals: Synthetic Minority Oversampling (SMOTE) Technique
Domino Data Lab
MAY 20, 2021
This renders measures like classification accuracy meaningless. Figure 3 shows visual explanation of how SMOTE generates synthetic observations in this case. note that this variant “performs worse than plain under-sampling based on AUC” when tested on the Adult dataset (Dua & Graff, 2017). References. link] Chawla, N.
Let's personalize your content