Pattern recognition based on machine learning identifies oil adulteration and edible oil mixtures

Area of Science:

• Food Science
• Analytical Chemistry
• Machine Learning

Background:

• Edible oils have unique fatty acid profiles, but these can be mimicked by mixtures, enabling fraudulent adulteration.
• Current methods struggle to identify oil types solely based on fatty acid composition, posing risks to food safety and public health.

Purpose of the Study:

• To develop a machine learning (ML) method for identifying ten distinct plant oil types based on their fatty acid profiles.
• To create a generalized supervised learning model capable of analyzing oil composition in mixtures and detecting adulteration.

Main Methods:

• Utilized a machine learning approach to identify discriminative fatty acid patterns across ten plant oil types.
• Developed a supervised end-to-end learning model trained on simulated oil mixtures for generalization.
• Incorporated on-line training capabilities to refine the deep learning model with new data and geographical variations.

Main Results:

• The ML model demonstrated high accuracy in identifying oil types and their intra-variability.
• Independent validation showed a 50th percentile absolute error of 1.4-1.8% and a 90th percentile error of 4-5.4% for 3-way oil mixtures.
• The model's ability to generalize and adapt through on-line training was confirmed.

Conclusions:

• The developed machine learning method offers a robust solution for identifying edible oil types and detecting adulteration.
• This technology can enhance product quality control, ensure fair pricing, and provide accurate labeling for health-conscious consumers.
• The findings pave the way for improved food safety and consumer trust in the edible oil market.