How does Naïve Bayes integrate with probabilistic calibration under temperature scaling?

Temperature scaling adjusts Naïve Bayes logits to improve probability calibration without changing decision boundaries.

Naïve Bayes often produces overconfident probabilities. Temperature scaling rescales logits by a factor T, softening or sharpening probability distributions. This improves calibration while preserving ranking. It is widely used in production ML systems for risk-sensitive predictions.