3. Model-Independent Preprocessing

An easy way to make existing cost-insensitive models cost-sensitive is to preprocess the data. Empulse provides a number of preprocessing techniques that can be used to make models cost-sensitive.

The first stream takes a page out of fairness literature and uses bias-mitigation techniques to remove the bias against a subgroup. This subgroup can strategically defined to serve your business needs. For instance, in a customer churn case, the subgroup can be defined as customers with high customer lifetime value (CLV). By removing the bias, you model will target more high-CLV customers.

The second stream uses sampling techniques based on the cost matrix.

• 3.1. Bias Mitigation
  • 3.1.1. Relabeling
  • 3.1.2. Resampling
  • 3.1.3. Reweighing
  • 3.1.4. References
• 3.2. Cost-Proportionate Sampling
  • 3.2.1. Rejection Sampling
  • 3.2.2. Oversampling
  • 3.2.3. Outlier Robustness
  • 3.2.4. Using the Cost-Sensitive Sampler in a Pipeline
  • 3.2.5. References