Support Vector Regression for the Determination of the Nutritional Components of Edible Oil by Terahertz Spectroscopy

Terahertz time-domain spectroscopy (THz-TDS) is used to investigate the principal nutritional component content in edible oil. For this purpose, different edible oils have been measured in the frequency range from 0.2 to 2.0 THz. The support vector regression (SVR) model that directly uses the raw terahertz spectral data as input data of the principal nutritional component is established and employed to determine principal component quantitative analysis for the THz-TDS. The predicted error of the four kinds of principal nutritional component (such as hexadecanoic acid, octadecanoic acid, octadecenoic acid, and octadecadienoic acid) content in edible oil is near 0.5%. Comparing the experimental data acquired by means of gas chromatography with predictions based on the SVR model exhibits excellent agreement and clearly indicates that SVR is a good candidate for the determination and prediction of the principal nutritional component content in edible oil on THz-TDS.