Review machine learning algorithms: Bayesian classifier

Naive Bayesian algorithm is to look for a great posteriori hypothesis (MAP), which is the maximum posteriori probability of the candidate hypothesis.

As follows:

In Naive Bayes classifiers, it is assumed that the sample features are independent from one another:

Calculate the posterior probability of each hypothesis and choose the maximum probability, and the corresponding category is the result of the sample classification.

Advantages and Disadvantages

Very good for small-scale data, suitable for multi-classification tasks, suitable for incremental training. At that time, it was necessary to have a high degree of independence between the characteristics of samples and not too much correlation. is sensitive to the form of input data expression.

Also, when a feature in a sample has a number of 0 occurrences in that category, that is, p (ai |vj) = 0, the numerator of the upper formula is all 0. This time requires the use of M-Estimator and Bayesian binding, as follows:

Review machine learning algorithms: Bayesian classifier