Support of data sets that contain missing values

Naive Bayes uses the values of each attribute only for conditional probability calculations, whereas it is assumed that the attributes of the class are conditionally independent. Therefore, missing values are ignored for each instance and each attribute during model creation and prediction.

More precisely, when a conditional attribute-value with the class probabilities of for each attribute , value , and class is estimated, only instances with non-missing values of are taken into account.

Similarly, when the product of such probabilities is calculated during the prediction for an instance , as shown in the following example, factors that correspond to the missing are skipped.

For data sets without missing values, the behavior remains unchanged.