Highly accurate prediction of edible oil oxidative stability by a multivariate regression model combining open ingredient data and chemiluminescence measurement

The oxidative stability of edible oils is mainly affected by fatty acid compositions and antioxidant contents. Open data on the ingredients of various edible oils is available, and predictions of oxidative stability using this data have been conducted. However, the accuracy of these predictions is insufficient. This is because the oxidation mechanism of edible oils is a complex system involving radical chain reactions. In this research, to enhance the prediction accuracy using a simple method, a multivariate regression model was constructed by incorporating new information on chemiluminescence (CL), which is related to the radical generation amount, in addition to the open ingredient data.

Thirteen kinds of commercial oils were selected to ensure a broad distribution of fatty acid compositions and antioxidant contents. The ingredient data of each oil were obtained from Standard Tables of Food Composition in Japan. As an index of oxidative stability, the induction period (IP) measured by the Rancimat method was used. When the CL of each oil was newly measured, two distinct types of emission patterns were observed. Therefore, explanatory variables for CL were set to reflect these differences in emission patterns.

To avoid complicating the method, the number of variables in the prediction model was limited to five. The prediction accuracy was evaluated using the adjusted coefficient of determination, r². In the linear regression analysis for IP prediction, the highest value of r² reached 0.946 in the model that included CL data, compared to the maximum of 0.877 in the model using only conventional ingredient data. Thus, the addition of CL information alone significantly improved the model accuracy.