Which distribution is commonly assumed for continuous features in Naive Bayes?

In Naive Bayes classification, when dealing with continuous features, the assumption that these features follow a Normal (Gaussian) distribution is commonly made. This assumption simplifies the process of estimating probabilities and classification since the normal distribution is characterized by its bell-shaped curve, defined by two parameters: the mean and the variance.

When applying Naive Bayes, if a feature is assumed to be normally distributed, the likelihood of that feature, given the class, can be computed using the probability density function of the normal distribution. This allows the algorithm to effectively model and make predictions based on continuous data, leveraging the properties of normal distribution, such as symmetry around the mean.

Other distributions mentioned may be applicable to different types of data or models but do not suit the common assumptions in context to continuous features in Naive Bayes. For instance, the Bernoulli distribution is typically used for binary variables, while the Poisson distribution is suitable for count data. The uniform distribution, which implies all outcomes are equally likely, does not provide the necessary detail for modeling continuous features effectively in a Naive Bayes context.