Model Interpretability

> A logistic model assigns one applicant a 4.6% approval score. A threshold policy turns that score into a decision, while the linear predictor can be decomposed exactly in the model's feature space.