EEG feature extraction and machine learning for consumer preference analysis of whisky.

Understanding consumer preference for food and beverages is central to product development, yet subjective self-reports alone often lack objectivity. Electroencephalography (EEG) can reveal implicit neural signals of preference. However, existing flavor studies utilizing EEG are mostly confined to group difference analysis, cannot identify stable preference-related EEG markers or achieve objective preference prediction. To address this gap, this study integrated sensory evaluation, systematic EEG feature extraction and machine learning to quantify and predict consumer whisky preference. Twenty-four panelists were tested with 12 whisky samples via 9-point hedonic and 5-point JAR scales. Time-domain and frequency-domain features were extracted from EEG recordings, and four classification models were trained to predict preference. Compared with the accuracy of models built on raw EEG data (≈51%), feature extraction significantly boosted performance, with the Random Forest model achieving the highest test accuracy of 82.6%. Feature weight analysis further identified potentially predictive EEG indicators, including frontal α asymmetry and α band power. This work verifies the predictive capacity of EEG features for machine learning modeling, thereby establishing a neurophysiologically based method for the objective evaluation. This scheme complements traditional subjective sensory assessment and delivers actionable technical guidance for industrial flavor tuning and product development. However, the high homogeneity of participants in age and background limits the generalizability of our experimental results.