摘要：Despite the unreasonable feature independence assumption, the naiveBayes classifier provides a simple way but competes well with more sophisticatedclassifiers under zero-one loss function for assigning an observation to a class giventhe features observed. However, it has been proved that the naive Bayes workspoorly in estimation and in classification for some cases when the features arecorrelated. To extend, researchers had developed many approaches to free of thisprimary but rarely satisfied assumption in the real world for the naive Bayes. In thispaper, we propose a new classifier which is also free of the independenceassumption by evaluating the dependence of features through pair copulasconstructed via a graphical model called D-Vine tree. This tree structure helps todecompose the multivariate dependence into many bivariate dependencies and thusmakes it possible to easily and efficiently evaluate the dependence of features evenfor data with high dimension and large sample size. We further extend the proposedmethod for features with discrete-valued entries. Experimental studies show thatthe proposed method performs well for both continuous and discrete cases.
