Tuning into flavor: predicting coffee sensory attributes from EEG with boosted-tree regression models

Area of Science:

• Neuroscience
• Food Science
• Machine Learning

Background:

• Flavor perception significantly influences consumer choices in products like coffee.
• Traditional sensory analysis methods can be time-consuming and subjective.

Purpose of the Study:

• To explore the efficacy of electroencephalography (EEG) combined with machine learning (ML) for predicting coffee's sensory attributes.
• To assess the potential of brainwave data in understanding flavor perception.

Main Methods:

• Extracted spectral and temporal features from EEG data of a professional panel tasting coffee.
• Employed Least-Squares Boosted Trees (LSBoost) models optimized via Bayesian hyperparameter tuning.
• Utilized a Leave-One-Subject-Out (LOSO) cross-validation scheme for robust evaluation.

Main Results:

• Achieved high predictive accuracy (Mean Absolute Error < 0.75) and robustness (Cohen's d > 0.6).
• Identified spectral powers and Hjorth's parameters in specific brain regions as key predictive features.
• Demonstrated superior performance compared to benchmark regression models.

Conclusions:

• EEG-based ML models show significant potential as an objective and efficient alternative to traditional Descriptive Sensory Analysis (DSA).
• This approach can provide valuable insights into the neural correlates of flavor perception for coffee and potentially other food products.