Determination of gossypol content in cottonseeds by near infrared spectroscopy based on Monte Carlo uninformative variable elimination and nonlinear calibration methods

Near infrared (NIR) spectroscopy combined with Monte Carlo uninformative variable elimination (MC-UVE) and nonlinear calibration methods employed to determine gossypol content in cottonseeds were investigated. The reference method was performed by high performance liquid chromatography coupled to an ultraviolet detector (HPLC-UV). MC-UVE was employed to extract the effective information from the full NIR spectra. Nonlinear calibration methods were applied to establish the models compared with the linear method. The optimal model for gossypol content was obtained by MC-UVE-WLS-SVM, with root mean squares error of prediction (RMSEP) of 0.0422, coefficient of determination (R2) of 0.9331, and residual predictive deviation (RPD) of 3.8374, respectively, which was accurate and robust enough to substitute for traditional gossypol measurements. The nonlinear methods performed more reliable than linear method during the development of calibration models. Furthermore, MC-UVE could provide better and simpler calibration models than full spectra.