User Guide

• 1. Introduction
  • 1.1. API’s of imbalanced-learn samplers
  • 1.2. Problem statement regarding imbalanced data sets
• 2. Over-sampling
  • 2.1. A practical guide
    • 2.1.1. Naive random over-sampling
    • 2.1.2. From random over-sampling to SMOTE and ADASYN
    • 2.1.3. Ill-posed examples
    • 2.1.4. SMOTE variants
  • 2.2. Mathematical formulation
    • 2.2.1. Sample generation
    • 2.2.2. Multi-class management
• 3. Under-sampling
  • 3.1. Prototype generation
  • 3.2. Prototype selection
    • 3.2.1. Controlled under-sampling techniques
      • 3.2.1.1. Random under-sampling
      • 3.2.1.2. Mathematical formulation
    • 3.2.2. Cleaning under-sampling techniques
      • 3.2.2.1. Tomek’s links
      • 3.2.2.2. Editing data using nearest neighbours
        • 3.2.2.2.1. Edited nearest neighbours
        • 3.2.2.2.2. Repeated Edited Nearest Neighbours
        • 3.2.2.2.3. All KNN
      • 3.2.2.3. Condensed nearest neighbors
        • 3.2.2.3.1. One Sided Selection
    • 3.2.3. Additional undersampling techniques
      • 3.2.3.1. Instance hardness threshold
• 4. Combination of over- and under-sampling
• 5. Ensemble of samplers
  • 5.1. Classifier including inner balancing samplers
    • 5.1.1. Bagging classifier
    • 5.1.2. Forest of randomized trees
    • 5.1.3. Boosting
• 6. Miscellaneous samplers
  • 6.1. Custom samplers
  • 6.2. Custom generators
    • 6.2.1. TensorFlow generator
    • 6.2.2. Keras generator
• 7. Metrics
  • 7.1. Classification metrics
    • 7.1.1. Sensitivity and specificity metrics
    • 7.1.2. Additional metrics specific to imbalanced datasets
    • 7.1.3. Macro-Averaged Mean Absolute Error (MA-MAE)
    • 7.1.4. Summary of important metrics
  • 7.2. Pairwise metrics
    • 7.2.1. Value Difference Metric
• 8. Cross validation
  • 8.1. Instance hardness and average precision
  • 8.2. Create imbalanced dataset with samples with large instance hardness
  • 8.3. Assess cross validation performance variance using InstanceHardnessCV splitter
• 9. Common pitfalls and recommended practices
  • 9.1. Data leakage
• 10. Dataset loading utilities
  • 10.1. Imbalanced datasets for benchmark
  • 10.2. Imbalanced generator
• 11. Developer guideline
  • 11.1. Developer utilities
    • 11.1.1. Validation Tools
    • 11.1.2. Deprecation
  • 11.2. Making a release
    • 11.2.1. Major release
    • 11.2.2. Bug fix release
• 12. References