Edited Naive Bayes
José María Martínez‐Otzeta, Basilio Sierra, Elena Lazkano, M. Ardaiz, Eduardo Jauregui · INTELIGENCIA ARTIFICIAL · 2006
"Naive Bayes is a well-known and studied algorithm both in statistis and mahine learning. Bayesian learning algorithms represent eah onept with a single probabilisti summary. This paper presents a variant of the Naive Bayes metho d, in whih the original training set is augmented in the following fashion: Leave-One- Out proedure is applied over the training set, and inorretly clasifed instanes aording to Naive Bayes model are dupliated. The augmented dataset is used to indue the model. The motivation behind this idea is that giving more importane to hard instanes (in this ase, dupliating them) might ontribute to make the model more aurate over that subset of the instane spae. We have tested this algorithm over 41 UCI datasets. The results suggest that the hane of obtaining a significant better performane than with the original Naive Bayes approah are muh greater than the opposite. "