Building a good model requires sufficient and balanced data. However, reality is often different. Rare disease diagnostic data has very few positive samples, fraudulent transactions represent only 0.1% of the total, and certain classes of images may have only a handful of examples. If this situation is left unchanged, the model will only match the majority class well and miss the important minority class.
Today, we cover the concepts of data augmentation and synthesis, as well as techniques for handling imbalanced data (such as SMOTE). The MLS-C01 exam asks not so much about code implementation, but rather "which technique is appropriate for which situation."