Classification of Consumer Preferences Using EEG Signals: A Machine Learning Approach in Neuromarketing Context

In this study, a machine learning approach was developed to classify consumer preferences through EEG signals. Using the open-access Yadava dataset, EEG recordings from 25 male participants aged 18-38 were analysed. Participants were shown different product images, and EEG signals were recorded for each product, along with feedback indicating "like" or "dislike." Data processing was performed using a 50 Hz notch, a 0.5-40 Hz bandpass, and Savitzky-Golay filters to remove noise. The EEG data was then segmented into five brain lobes (frontal, temporal, parietal, occipital, and all lobes) and five frequency bands (delta, theta, alpha, beta, and gamma). In the feature extraction phase, Hjorth parameters (activity, mobility, complexity), entropy-based metrics (Shannon, Tsallis, LogEnergy), statistical features (mean, variance, skewness, kurtosis, etc.), and power spectral density (PSD)-based spectral features were used. The resulting feature matrix was classified using SVM, KNN, NB, and RF algorithms, and 10-fold cross-validation was applied. According to experimental results, the RF algorithm demonstrated the highest performance with 99% accuracy, 100% precision, 99% recall, and 99% F1-score. Furthermore, the KNN algorithm achieved the lowest computational cost in terms of processing time. The study provides a suitable solution for real-time neuromarketing applications because it also considers the computational cost parameter, which is often neglected in real-time system integration. Future studies aim to further enhance emotional state and preference prediction by integrating different classification methods and brain connectivity analyses.

In this study, a machine learning approach was developed to classify consumer preferences through EEG signals. Using the open-access Yadava dataset, EEG recordings from 25 male participants aged 18-38 were analysed. Participants were shown different product images, and EEG signals were recorded for each product, along with feedback indicating "like" or "dislike." Data processing was performed using a 50 Hz notch, a 0.5-40 Hz bandpass, and Savitzky-Golay filters to remove noise. The EEG data was then segmented into five brain lobes (frontal, temporal, parietal, occipital, and all lobes) and five frequency bands (delta, theta, alpha, beta, and gamma). In the feature extraction phase, Hjorth parameters (activity, mobility, complexity), entropy-based metrics (Shannon, Tsallis, LogEnergy), statistical features (mean, variance, skewness, kurtosis, etc.), and power spectral density (PSD)-based spectral features were used. The resulting feature matrix was classified using SVM, KNN, NB, and RF algorithms, and 10-fold cross-validation was applied. According to experimental results, the RF algorithm demonstrated the highest performance with 99% accuracy, 100% precision, 99% recall, and 99% F1-score. Furthermore, the KNN algorithm achieved the lowest computational cost in terms of processing time. The study provides a suitable solution for real-time neuromarketing applications because it also considers the computational cost parameter, which is often neglected in real-time system integration. Future studies aim to further enhance emotional state and preference prediction by integrating different classification methods and brain connectivity analyses.