Simple bayesian classifier applied to learning

Abstract

In this article, we propose the use of a new simple Bayesian classifier (SBND) that quickly learns a Markov boundary of the class variable and a network structure relating class variables and the said boundary. This model is compared with other Bayesian classifiers, then experimental tests are carried out for which 31 well-known ICU databases and two bases of artificial variables have been used. With these databases we compare the results obtained by such algorithms studied in the state of the art such as Naive Bayes, TAN, BAN, RPDag, CRPDag, SBND and combinations with different metrics such as K2, BIC, Akaike, BDEu. The experimental work was done in Elvira software.