A COPULA-BASED SUPERVISED LEARNING CLASSIFICATION FOR CONTINUOUS AND DISCRETE DATA
Abstract
The naive Bayes classifier offers a straightforward method for allocating an observation to a
class given the characteristics seen, yet it competes effectively with more advanced classifiers under the
zero-one loss function despite the irrational feature independence requirement. However, it has been shown
that when the features are correlated, the naive Bayes performs badly in both estimation and classification.
In order to overcome this fundamental but seldom met assumption for the naive Bayes, researchers have
created a variety of strategies. In this research, we present a novel classifier that does not rely on the
independence assumption by assessing feature dependency using pair copulas built using a graphical model
known as the D-Vine tree. Even for data with high dimensions and huge sample sizes, this tree form
facilitates the easy and fast evaluation of feature dependency by breaking down the multivariate dependence
into several bivariate dependencies. For features with discrete-valued entries, we expand the suggested
approach. The suggested approach works effectively in both continuous and discrete situations, according
to experimental research.