Novel Methodology to Improve the Accuracy of Oxide Determination in Cement Raw Meal by near Infrared Spectroscopy (NIRS) and Cross-Validation-Absolute-deviation-F-Test (CVADF)

Near infrared spectroscopy (NIRS) combined with a partial least squares (PLS) algorithm was utilized as a rapid alternative analytical method for estimating the four main oxides in cement raw meal samples. An algorithm, known as cross-validation-absolute-deviation-F-test (CVADF), was proposed to eliminate the outliers existed in the calibration set. 5, 6, 2 and 2 out of 76 samples were identified as outliers for CaO, SiO2, Al2O3 and Fe2O3, respectively. The correlation coefficient of prediction (Rp) increased from 0.7773, 0.7877, 0.8894 and 0.6357 to 0.9075, 0.8572, 0.9038 and 0.6400, while the root mean square error of prediction (RMSEP) decreased from 0.2493, 0.2331, 0.0832 and 0.0449 to 0.1664, 0.1949, 0.0779 and 0.0447, respectively, indicating that the outliers are accurately identified and that the prediction performance of the PLS models established by the remaining samples was significantly improved. Some common outlier elimination methods, leverage diagnostic (LD), Euclidean distance diagnostic (EDD), Mahalanobis distance diagnostic (MDD) and principal component scores diagnostic (PCSD) were used for comparison. The results show that the proposed method is very promising with good results for the prediction capability.