Stability evaluation of Neural and statistical Classifiers based on Modified Semi - bounded Plug - in Algorithm

This paper illustrates a new criterion for evaluating neural networks stability compared to the Bayesian classifier. The stability comparison is performed by the error rate probability densities estimation using the modified semi-bounded Plug-in algorithm. We attempt, in this work, to demonstrate that the Bayesian approach for neural networks improves the performance and stability degree of the classical neural classifiers. Keywords—Bayesian neural networks, error rate density, modified semi-bounded Plug-in algorithm, stability, theoretical error rates.