The use of Bayesian Networks in the prediction of bankruptcy

Abstract

The purpose of this paper is to compare the development of Bayesian Network models: naïve
Bayesian network (NBN) and a maximum weight spanning tree (MWST) in forecasting the
failure of the Tunisian companies. Using a sample of 130 small and medium Tunisian
companies and a battery of 9 financial ratios calculated for the 2005-2012 period, it can be
concluded that, for a classification problem, supervised learning with naïve architecture is
more appropriate and gives more relevant results than an unsupervised learning model with a
maximum weight spanning tree.