Predict first: A query is moderately compatible with class A on both features, but extremely compatible with class B on only one feature. Which class wins after multiplying the evidence?
Move the query point
P(A | x)—
P(B | x)—
Lecture 8 · Continuous features
Each continuous feature has a class-conditional Gaussian distribution. Under the Naïve Bayes assumption, the per-feature likelihoods multiply.