Self-organizing maps and learning vector quantization networks as tools to identify vegetable oils.

Area of Science:

• Analytical Chemistry
• Machine Learning in Food Science

Background:

• Edible and vegetable oil authentication is crucial for consumer safety and regulatory compliance.
• Detecting adulteration of extra virgin olive oil (EVOO) requires robust analytical methods.
• Fatty acid profiles are key indicators for oil identification and adulteration detection.

Purpose of the Study:

• To evaluate the efficacy of Self-Organizing Map (SOM) and Learning Vector Quantization (LVQ) models for identifying edible and vegetable oils.
• To assess the capability of SOM and LVQ models in detecting adulteration of EVOO with common vegetable oils.
• To determine the minimum detectable concentrations of adulterants in EVOO using the developed models.

Main Methods:

• Utilized Self-Organizing Map (SOM) and Learning Vector Quantization (LVQ) network models.
• Trained and validated models using bibliographical databases for internal validation.
• Performed external validation using six distinct databases.
• Analyzed model performance based on misclassification rates.

Main Results:

• SOM and LVQ models demonstrated high classification accuracy, exceeding 94% in the worst-case scenarios.
• Successfully detected adulteration of EVOO with corn, soya, sunflower, and hazelnut oils.
• Established detection thresholds for adulterants: corn oil >10%, soya oil >5%, sunflower oil >5%, and hazelnut oil >10%.

Conclusions:

• SOM and LVQ models are effective tools for the identification of edible oils and the detection of EVOO adulteration.
• The models provide reliable quantitative insights into the minimum concentrations of adulterants detectable.
• These machine learning approaches offer a promising avenue for quality control in the edible oil industry.