liveBook · Manning

SMOTE

As explained in Build AI Drug Discovery Pipelines

SMOTE is an oversampling technique that creates synthetic minority-class instances along line segments connecting existing minority-class instances.

SMOTE addresses class imbalance by generating synthetic minority-class instances along lines connecting existing minority-class instances, rather than duplicating them. section 5.2.2

Use SMOTE as one option when imbalanced training data could steer classification learning off track:

  • Compared with random oversampling, it avoids simply repeating the same minority instances, which can contribute to overfitting. section 5.2.2
  • Compared with random undersampling, it does not discard majority-class instances, though sampling strategies have different trade-offs. section 5.2.2

Practical use

Apply SMOTE through the imblearn library; compare its training-data class composition with downsampling and upsampling. section 5.2.2

Figure 5.8 Comparison of downsampling and upsampling on the training data set’s class composition.

Inspect the class-composition comparison for downsampling and upsampling on training data. section 5.2.2

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Defines synthetic-instance generation and contrasts sampling trade-offs, while deferring advanced SMOTE variants.
Examples
Uses a training-set class-composition figure and recommends the imblearn library for applying SMOTE.
What it assumes
Recognizes majority and minority classes and understands that imbalanced training data can distort classification learning.
What you can do after
Apply it via the imblearn library to generate minority-class training instances without simple duplication.
How it connects
Connects SMOTE with class imbalance, random oversampling, random undersampling, synthetic data generation, and overfitting.

Also explained in

Compare how 2 books explain SMOTE →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.