The use of data set reliability factors in a multidimensional Bayesian classifier

This paper evaluates the reliability factors influence in a multidimensional Bayesian classifier. To evaluate the best reliability factors values, an accuracy index based on five well known accuracy coefficients was proposed. The results for a two-dimensional classifier show that, instead of a unique combination of reliability factors that achieves the best results, there is a linear relation between the best choices of reliability factors. A method for the reliability factors estimation is also proposed.