Machine Learning Assisted Analysis of Relationships between Trans Fatty Acid Content and Fatty Acid Composition in Vegetable Oils and Margarines

Abstract

Objective: Trans fatty acid (TFA) intake raises public health concern owing to their potential adverse effects on health. Reduced intake is thus highly recommended. We aimed to identify indicators for selecting vegetable oils and margarines with low TFA contents by analyzing their association with fatty acid composition using machine learning.

Methods: TFA content and fatty acid composition in 32 vegetable oil and 28 margarine samples were measured using gas chromatography. Ridge regression and decision tree analyses were performed to predict and classify TFA content. For vegetable oils, fatty acid composition was used as the explanatory variable, and both fatty acid composition and lipid content were used for margarines.

Results: Respective average TFA contents in vegetable oils and margarines were 0.41 g/100 g and 0.43 g/100 g. Ridge regression was used to construct predictive models for TFAs in vegetable oils and margarines, revealing significant correlations between measured and predicted values. Decision tree analysis indicated that vegetable oils with less than 13% diene fatty acids, and margarines with less than 4.1% triene fatty acids can be classified as products with low TFA content.

Conclusions: These results suggest that vegetable oils with proportionally low diene fatty acid content and margarines with proportionally low triene fatty acid content tend to have low TFA contents.