Hands-On Data Analysis with Pandas

Over-sampling

It's clear that with smaller datasets, it won't be beneficial to under-sample. Instead, we can try over-sampling the minority class (the high-quality red wines, in this case). Rather than doing random over-sampling with the RandomOverSampler, we are going to use the Synthetic Minority Over-sampling Technique (SMOTE) to create new (synthetic) red wines similar to the high-quality ones using the k-nearest neighbors algorithm. By doing this, we are making a big assumption that the data we have collected about the chemical properties of the red wine does influence the quality rating of the wine.

The SMOTE technique in imblearn comes from this paper: N. V. Chawla, K. W. Bowyer, L. O.Hall, W. P. Kegelmeyer, SMOTE: synthetic minority ...