In a regression problem, the ratio of the prediction error standard deviation to the original output data standard deviation. A lower S.D. ratio indicates a better prediction. This is equivalent to one minus the explained variance of the model. See Multiple Regression, Neural Networks.