Characterizing Edible Oils by Oblique-Incidence Reflectivity Difference Combined with Machine Learning Algorithms.

Area of Science:

• Analytical Chemistry
• Food Science
• Machine Learning

Background:

• Edible oil adulteration, driven by price disparities, is a significant consumer concern.
• Adulterated oils pose risks to consumer health and well-being.
• Current quality testing methods lack speed, non-destructiveness, or universal applicability.

Purpose of the Study:

• To introduce a fast, non-destructive, accurate, and reliable method for edible oil quality testing.
• To detect various types of edible oil adulteration using a novel approach.
• To evaluate the effectiveness of machine learning algorithms in conjunction with a specific optical technique.

Main Methods:

• Utilized the oblique-incidence reflectivity difference (OIRD) method for optical measurements.
• Integrated machine learning algorithms including Gradient Boosting, K-Nearest Neighbor, and Random Forest for data analysis.
• Analyzed the contribution of OIRD signal, DC signal, and fundamental frequency signal to classification.

Main Results:

• Machine learning models achieved prediction accuracies exceeding 95% for edible oil detection.
• The OIRD signal demonstrated the highest contribution rate (45.7%) to classification.
• In olive oil quality evaluation, the OIRD signal showed significantly higher feature importance (63.4%) compared to DC and fundamental frequency signals.

Conclusions:

• The oblique-incidence reflectivity difference (OIRD) method is a powerful tool for detecting edible oils and their adulteration.
• Machine learning algorithms effectively enhance the accuracy and reliability of OIRD-based oil analysis.
• The OIRD method provides a promising solution for rapid, non-destructive quality control in the edible oil industry.