Logistic Regression

Interpretation

• Numerical Feature: If we increase the value of feature by one unit, the estimated odds change by a factor of exp(weight). Final weight = weight * exp(weight)
• Categorical Feature: One of the value of a feature is reference category. Changing the value of feature from reference category to other category will change estimated odds by a factor of exp(weight)
• Intercept: When all numeric features are zero and categorical features are at reference category, the extimated odds are exp(bias). Intercept weight is ususally not relevant

Pros and Cons

• Same as linear regression
• Interpretation is more difficult as the interpretation of weights is multiplicative and not additive