RPD function

Residual Prediction Deviation (RPD)

This function computes the Residual Prediction Deviation (RPD), which is defined as the standard deviation of observed values divided by the Root Mean Square Error or Prediction (RMSEP). The RDP takes both the prediction error and the variation of observed values into account, providing a metric of model validity that is more objective than the RMSEP and more easily comparable across model validation studies. The greater the RPD, the better the model's predictive capacity.

RPD(predicted, observed, na.rm = FALSE)

Arguments

• predicted: a numeric vector containing predicted values.
• observed: a numeric vector of the same length as predicted containing observed values.
• na.rm: Boolean parameter indicating whether NA values should be removed before the analysis

Returns

numeric value of the RDP.

Details

Interpretation of the RPD is somewhat arbitrary, with different thresholds for a good model used in the literature. Many studies call a model excellent, when the RPD is above 2 (but other classification use thresholds as high as 8 for this).

Examples

predicted<-c(1,2,3,4,5,6,7,8,9,10)
observed<-c(1.5,1.8,3.3,3.9,4.4,6,7.5,9,11,10)

RPD(predicted,observed)

References

Williams PC and Sobering DC (1993) Comparison of commercial near infrared transmittance and reflectance instruments for analysis of whole grains and seeds. J. Near Infrared Spectrosc. 1, 25-32 (I didn't have access to this paper, but have noticed that it is often provided as the key reference for the RPD).

Author(s)

Eike Luedeling