Liking State Estimation Using Time-Frequency Image Representation of EEG Signals

Abstract

In recent years, there has been a significant increase in research on emotion and preference state estimation. In this study, a preference prediction method is proposed for use in neuromarketing studies by utilizing time-frequency (TF) energy distribution images derived from electroencephalogram (EEG) signals. EEG signals recorded while participants watched commercials from two different automobile brands were evaluated using deep learning techniques to estimate their preference states. After viewing the advertisements, participants were shown selected visual segments from the commercials (e. g., front view, dashboard, etc.) and asked to rate their preferences on a scale from 1 to 5. The EEG signals corresponding to these segments were transformed into two-dimensional RGB-scaled scalogram/spectrogram images using Continuous Wavelet Transform (CWT) and Short-Time Fourier Transform (STFT). Using deep learning models, the proposed method achieved maximum classification accuracies of 86.61% and 87.26%, respectively.