Prediction of nitrophenol-type compounds using chemometrics and spectrophotometry

Two chemometric methods, WPT-ERNN and least square support vector machines (LS-SVM), were developed to perform the simultaneous spectrophotometric determination of nitrophenol-type compounds with overlapping spectra. The WPT-ERNN method is based on Elman recurrent neural network (ERNN) regression combined with wavelet packet transform (WPT) preprocessing and relies on the concept of combining the idea of WPT denoising with ERNN calibration for enhancing the noise removal ability and the quality of regression without prior separation. The LS-SVM technique is capable of learning a high-dimensional feature with fewer training data and reducing the computational complexity by requiring the solution of only a set of linear equations instead of a quadratic programming problem. The relative standard errors of prediction (RSEPs) obtained for all components using WPT-ERNN, ERNN, LS-SVM, partial least squares (PLS), and multivariate linear regression (MLR) were compared. Experimental results showed that the WPT-ERNN and LS-SVM methods were successful for the simultaneous determination of nitrophenol-type compounds even when severe overlap of spectra was present.