Fast and accurate determination of organophosphate pesticides using surface-enhanced Raman scattering and chemometrics

Abstract

We have developed an approach for the fast and accurate determination of organophosphate pesticides by combining surface-enhanced Raman scattering (SERS) technology with chemometric methods. We measured the SERS spectra of three pesticides (methyl parathion, edifenphos and ethyl paraxon) and then processed the spectra using the following pre-processing algorithms: baseline subtraction; Savitzky–Golay first derivative, standard normal variate variance (SNV) transformation and multiplicative scatter correction. Principal components analysis and non-negative matrix factorization were subsequently adopted to obtain the main features of the spectra. The data were then used to develop classification models by support vector machines and random forest (RF) regression and the discrimination performance was evaluated through a five-fold cross-validation method. The experiments showed that the baseline subtraction method can perfectly eliminate the fluorescence background and baseline drift. Principal components analysis greatly shortens the training time on the premise of maintaining classification accuracy, although non-negative matrix factorization gives poorer results. SNV can improve the discriminant accuracy up to 4%, except for the use of non-negative matrix factorization, but multiplicative scatter correction and the first derivative method have a negative effect. The classification model of the highest accuracy (99.79%) is built by support vector machines with the spectra processed by SNV and principal components analysis; and the training process takes 1115 s. The training time (81 s) of the RF model developed with the spectra processed by SNV is much shorter than the former method, although the accuracy is similar. The classification accuracy of the model built with RF and different data always maintains a high level, suggesting that RF has excellent robustness. These results show that the most suitable method for the determination of organophosphate pesticides is a combination of SERS spectrometry and RF with SNV.