Using neural networks committee machines to improve outcome prediction assessment in nonlinear regression

This study proposes a methodology to improve nonlinear regression model prediction assessment by the construction of a model for error pattern recognition to estimate whether the model outcome prediction value will fall outside the model confidence interval. The methodology was evaluated on six experiments using five widely known public databases from UCI Machine Learning Repository. The essays provided evidences that the pattern recognition models were able to identify observations, in the testing datasets, that are more likely to produce higher error values, and improve the overall outcome of the nonlinear regression models predictions.