Detection of prediction outliers and inliers in multivariate calibration Original Research Article

After a multivariate calibration model is built and validated, it is ready to be used for the prediction of the characteristic to be determined in the new samples. Before prediction, one should make sure that the new samples are similar to the calibration samples. If not, such samples are called prediction outliers, which can be of two types, namely samples that are outside of the calibration space, and samples which are situated in a low- (or empty-) density region of the calibration space. The latter type, also called inliers, are not detected by existing methods. In this paper, potential functions are proposed for the detection of inliers and outliers in prediction.