Outlier Detection for Multivariate Calibration in Near Infrared Spectroscopic Analysis by Model Diagnostics

Outlier detection is an important task in multivariate calibration because the quality of a calibration model is determined by that of the calibration data. An outlier detection method was proposed for near infrared (NIR) spectral analysis. The method was based on the definition of outlier and the principle of partial least squares (PLS) regression, i.e., an outlier in a dataset behaved differently from the rest, and the prediction result of a PLS model was an accumulation of several independent latent variables. Therefore, the proposed method built a PLS model with a calibration dataset, and then the contribution of each latent variable was investigated. Outliers were detected by comparing these contributions. An NIR spectral dataset of orange juice samples was adopted for testing the method. Six outliers were detected in the calibration set. The root mean squared error of cross validation (RMSECV) was reduced from 16.870 to 4.809 and the root mean squared error of prediction (RMSEP) was reduced from 3.688 to 3.332 after the removal of the outliers. Compared with a robust regression method, the result of the proposed method seemed more reasonable.

An outlier detection method was proposed for near infrared (NIR) spectral analysis based on diagnosing the contribution of the samples in each latent variable to the partial least squares (PLS) model. Local outlier factor (LOF) was used for discriminating the outliers.Figure optionsDownload as PowerPoint slide