Abstract: Outlier samples in near infrared spectroscopy analysis can strongly influence on the performance of the prediction model. To detect and eliminate the outlier samples, a new outlier sample eliminating algorithm base on joint XY distances (ODXY) was presented, and the relation of XY distances of NIR is proposed and proved. In this research, 102 lamb samples were collected and the data of NIR spectroscopy and moisture content was measured and analyzed. Initially, Mahalanobis distances method, Monte-Carlo sampling method and ODXY method to were employed to eliminate the outlier samples and built the PLS prediction model based on the processed samples. Then, the predictive mean square error (RMSEP) and the coefficient of determination (R2) were used to test the performance of the prediction model. Finally, the generalization of the eliminating algorithm was tested by new calibration and validation sets. The experiments show that ODXY method has better performance and better generalization ability than the other methods tested in our experiments.

Key words: outlier samples, prediction model, near infrared spectroscopy, Mahalanobis distance, Monte-Carlo sampling method