Robust local PCR and its application in NIR spectral quantitative analysis

Local principle component regression (PCR) is a widely-used nonlinear calibration method in near infrared (NIR) spectral quantitative analysis. However, it is more sensitive to outliers than global PCR. An algorithm which combines robust PCR and local regression is proposed in this paper. Local training samples are selected according to Mahalanobis distances and the two stages of PCR, principle component analysis (PCA) and multivariate linear regression (MLR), are replaced with robust minimum covariance determinant (MCD) estimation and least median squares (LTS) regression respectively. The proposed algorithm is applied to a set of gasoline samples and experimental result shows that the predictive root mean square error of the proposed algorithm is obviously smaller than global robust PCR while their breakdown points comparable.