Perturbation Effect on Logistic Regression Predictions
Problem statement
How much does a small feature change move a logistic regression prediction? For $\hat p=\sigma(\beta^T X)$, where $\sigma(z)=1/(1+e^{-z})$, increase only $X_j$ by $\delta$ and approximate the probability change. Then explain how the sensitivity depends on the current prediction.
Show hint
Which scale is the model linear on, and how does the nudge change the quantity on that scale?
Relevant roles: Data Scientist, Applied Scientist and ML Engineer.
Related free problems
- Logistic Regression: Gradient, Hessian, and ConvexityStatistics & InferenceMedium