PLS model investigation of thiabendazole based on THz spectrum

Abstract THz spectrum of thiabendazole was determined with terahertz time domain spectroscopy (THz-TDS) at room temperature, and comparison of its THz spectrum with other two bactericides (thiophanate-methyl and benomyl) was given. The solid-state density functional and isolated-molecule theories were employed to interpret thiabendazole's THz spectrum with the PW91 density function and d and p polarization (DNP) basis set. In addition, we applied THz-TDS to detect mixtures of thiabendazole and polyethylene with different ratios, and compared the performances of four variable selection methods for the prediction of thiabendazole content. The tested variable selection algorithms included partial least squares (PLS), interval PLS (iPLS), backward iPLS (biPLS) and moving window PLS (mwPLS). The PLS model intervals were optimized according to the root mean square error of cross validation (RMSECV) in the calibration set. The performances of models were evaluated according to the root mean square error of prediction (RMSEP), correlation coefficient ( R ) and relative error ( E ) in the prediction set. In contrast with the other three methods, the mwPLS yielded the most accurate result. The optimal mwPLS model was obtained with RMSEP=0.3629, R =0.9995 and E =3.75%, and the investigation of mwPLS models with various window sizes from 3 to 71 were examined.