Nondestructive detection of frying times for soybean oil by NIR-spectroscopy technology with Adaboost-SVM (RBF)

Soybean oil with fried repeatedly at high temperature produces harmfully substances to threaten human health. In this paper, Adaboost-SVM (RBF) classification model combined with near infrared spectroscopy (NIRS) was proposed to detect frying times of soybean oil based on the optimal selection of crucial parameters and the combination of Adaboost, support vector machine (SVM) of radial basis function (RBF) as the kernel function. Herein, four modes were designed to divide 15 fryings into primary and secondary stages, then the created classification models were compared. Specially, after classifying as I, II, III class by mode 3, the accuracy of primary model established by Adaboost reached 98 %. Afterwards, spectra with pre-processed by various methods were analyzed. As expected that the performance of secondary models were significantly reinforced, especially the accuracy of SVM (RBF) model increased from 76 % to 88.89 %. Additionally, successive projections algorithm (SPA) was applied to obtain relevant wavelengths. When the dimensions of I, II, III class were severally reduced to 14, 6, 8, the accuracy of SVM (RBF) model was the best average value of 93.33 %. Finally, the results of external validation indicated that the accuracy of primary and secondary models reached 95.55 %, 91.11 %, respectively.