What is probabilistic calibration drift in Naïve Bayes over time?

Calibration drift occurs when Naïve Bayes probability outputs become misaligned with true observed frequencies over time.

Even if classification accuracy remains stable, probability estimates can drift due to evolving feature distributions or violated independence assumptions. This leads to overconfident or underconfident predictions. Calibration techniques like Platt scaling or isotonic regression are used to correct drift in production systems.