Application of principal component analysis‐multivariate adaptive regression splines for the determination of surface area in zeolites

Abstract

AbstractIn this study, the principal component analysis‐multivariate adaptive regression splines (PCA‐MARS) technique was applied to efficiently predict the surface area in zeolites. Before building the MARS model, the data matrix obtained from infrared spectroscopy was subjected to robust principal component analysis (rPCA) and six principal component (PC) scores with the explained validation variance 97.983% were used as inputs to the MARS algorithm. The training set (72 × 6) was used to develop the rPCA‐MARS model, and the efficiency of the proposed method was evaluated in terms of coefficient of determination (R2), R2 estimated by generalized cross‐validation (R2GCV), root mean square error of calibration (RMSEC), root mean square error of prediction (RMSEP), and mean absolute error (MAE). R2, RMSEC, RMSEP, MAE, and R2GCV in the piecewise‐linear rPCA‐MARS model were 0.9976, 1.4166, 1.8370, 1.0120, and 0.9955, respectively. We proposed also to use varimax (Var) rotated of the significant principal components as input to the MARS algorithm (R2 = 0.9964, RMSEC = 1.4214, RMSEP = 1.9031, MAE = 1.0511, and R2GCV = 0.9951). Upon viewing the PCA‐Var‐MARS results, we concluded that the obtained results are comparable with rPCA‐MARS results. In this study, principal component regression (PCR), partial least squares regression (PLSR), and support vector machine regression (SVMR) methods were also used for the quantitative determination of the surface area of zeolites. The RMSEP for PCR, PLSR, and SVMR was 11.5636, 8.7111, and 7.2426, respectively. Based on the results obtained, the MARS model is more reliable than commonly used regression models. It can be concluded that the rPCA‐MARS method is an appropriate method for analyzing spectral data to determine the textural properties of zeolite.