The Support Vector Classification (SVC) model was constructed in this study by applying the near infrared (NIR) analysis technology combined with the chemometrics method to classify and distinguish 11 types of vegetable oil, including "restaurant waste oil". Partial least squares-discriminate analysis (PLS-DA) and GA-SVC classification models were constructed, respectively, after the data were preprocessed with Savitzky-Golay (SG) smoothing, Standardized normal variate and De-trending (SNV-DT) and unit vector normalization (UVN). The results indicated that the pre-processing effect of SNV-DT was the best, PLS-DA failed to complete the classification, while GA-SVC model was able to make accurate prediction, the penalty parameter of SVC classifier C=642.5 was too large, and the model had poor generalizability and the modeling was time consuming. Back Interval Partial Least Squares (BiPLS), Successive Projection Algorithm (SPA), and BiPLS-SPA were adopted respectively to extract the characteristic wavelength to construct GA-SVC classification model. Of these three methods, the wavelength selected with BiPLS-SPA was 17, accounting for only 3.77% of the full wave, the penalty parameter was C=32.53, and the identification rates of prediction set and calibration set reached 95% and 99.4%, respectively. This model also had the shortest modeling time, and better generalizability. This research indicated that the method of NIR combined with GA-SVC could classify and distinguish edible vegetable oils rapidly, accurately and effectively. BiPLS-SPA could effectively reduce the collinearity among variables, allowing the model to be more robust with high generalizability.