Machine learning based classification of yogurt aroma types with flavoromics

Area of Science:

• Food Science
• Analytical Chemistry
• Computational Biology

Background:

• Traditional sensory evaluation of food products like yogurt is subjective and lacks automation.
• The complexity of human sensory perception presents a significant challenge for developing objective, computational alternatives.
• Objective methods are needed to ensure consistent quality and accurate characterization of food aromas.

Purpose of the Study:

• To develop a computational method for predicting yogurt aroma types based on chemical composition.
• To establish an automated pipeline for monitoring the sensory properties of yogurts.
• To identify key aroma-active compounds indicative of specific yogurt aroma profiles.

Main Methods:

• Construction of logistic regression classification models.
• Analysis of aroma-active compound concentrations as input features.
• Utilization of feature importance analysis to identify indicator compounds.
• Derivation of classification criteria for yogurt aroma types.

Main Results:

• Logistic regression models achieved high classification accuracy for predicting yogurt aroma types (AUC ROC > 0.8).
• Feature importance analysis successfully identified key indicator compounds responsible for differentiating aroma types.
• Classification criteria based on these indicator compounds were successfully derived.

Conclusions:

• Machine learning models provide an accurate and automated approach to classifying yogurt aroma types.
• The identified indicator compounds and derived criteria offer a robust method for objective sensory monitoring.
• This automated pipeline can enhance quality control and product development in the dairy industry.