Neural network ensembles for posterior probability estimation

Neural network classification models have been successfully demonstrated in many application areas. The theoretical ability of neural networks to estimate posterior probabilities help to explain these successes. This research investigates the use of neural network ensembles for posterior probability estimation. Neural ensembles combine estimates from several networks to improve model estimation. In addition to the basic ensemble method, advanced models incorporating the resampling techniques of bootstrapping and k-fold cross validation are investigated for simulated problems of realistic complexity and limited training sample sizes.