Qualitative and quantitative detection of peanut oils adulteration based on fatty acid information fusion coupled with chemometrics

Based on fatty acid data, a new rapid method combining data fusion and chemometrics was established to detect peanut oils (PNO) adulteration qualitatively and quantitatively. The datasets from overall and sn-2 fatty acid profiles of 350 PNO samples were constructed by serial, parallel, and triglyceride fusions. Linear discrimination analysis (LDA) and support vector machine classification (SVC) were used to identify PNO adulteration. The model parameters of SVC were optimized using the grid search method and the particle swarm optimization algorithm. Principal component regression, partial least squares, and support vector machine regression (SVR) were used to predict adulterant contents. Results showed that SVC-based models had better identification capability than LDA-based models. In qualitative terms, SVC-based model using serial fusion data achieved 100% identification accuracy and fidelity. In quantitative terms, SVR-based adulteration model was significantly better than PCR- and PLS-based ones. Compared with SVR models using overall and sn-2 fatty acid data, SVR-based model using triglyceride fusion data had a root mean square error of prediction of only 0.0513. Therefore, it is feasible to combine data fusion and chemometrics for detecting PNO adulteration. The method also provides theoretical foundation and technical support for the safety and quality detection of edible oils.