Optimal committee of probabilistic neural networks for statistical pattern recognition

In this paper, an optimal committee of probabilistic neural networks is designed based on mixture model algorithm. Kernel-based approaches are used to estimate the probability densities. The kernel size, h(i), is assumed to be a random variable. The unknown distribution of h(i), P(h(i)) is denoted as mixing parameters from mixture model, which is optimally trained based on maximum likelihood involving non-linear optimization and re-estimation. The committee of probabilistic neural network is applied to optimally estimate the density function of data space, thus the data sample drawn can be classified into the appropriate category using Bayesian decision rule.