Consumer Preference Prediction with mRMR-Based Explainable EEG Classification

Authors : Suzan Saban, Eda Dağdevir

Pages : 125-131

View : 79 | Download : 201

Publication Date : 2025-12-23

Article Type : Research Paper

Abstract :This study aimed to classify consumer taste using EEG signals. An open-access EEG dataset was used in the study, and a total of 1045 EEG recordings were obtained from 25 participants aged 18–38. Data were recorded with a 14-channel Emotiv Epoc+ device at a sampling frequency of 128 Hz. After preprocessing, a total of 1190 features were extracted from each channel based on time, entropy, statistics, and spectral data. The Minimum Redundancy Maximum Relevance (mRMR) algorithm was used for feature selection, and the six most informative features were identified. Support Vector Machines (SVM), K-Nearest Neighbor (KNN), Naive Bayes (NB), and Random Forest (RF) algorithms were applied during the classification phase, and model performance was evaluated using 10-fold cross-validation. In the classification performed with the full feature set, the RF algorithm achieved the highest accuracy rate, with 99%. Even using only six features selected using the mRMR method, the RF model achieved 95% accuracy, an F1 score of 95%, and a sensitivity of 94%. A significant contribution of the study is that, in addition to achieving high accuracy, it also increases the model\\\'s explainability by clarifying which EEG channel and frequency band each feature corresponds to. In this respect, the study provides an explainable artificial intelligence approach to EEG-based neuromarketing studies. In conclusion, achieving high accuracy and interpretability using a small number of features selected using the mRMR method represents a significant advance in EEG-based consumer taste prediction in terms of both computational efficiency and physiological interpretation.
Keywords : Nöropazarlama, EEG, Sinyal işleme, Makine Öğrenmesi, mRMR

ORIGINAL ARTICLE URL