Update a posterior probability from a prior and a likelihood using Bayes theorem. Applies to medical screening, spam filtering and any belief-updating task.
即使灵敏度和特异度都很高,低先验也会使后验较低。
| Item | Formula / Note | Source / Standard |
|---|---|---|
| Bayes rule | posterior = likelihood x prior / evidence | Bayesian inference |
| Naive assumption | naive Bayes assumes conditional independence at the cost of bias | Naive Bayes literature |
| Zero probability | unseen features zero the product unless smoothing is applied | Naive Bayes literature |