EEG Classification to Food Stimuli in Diverse Weight Groups with Regression Analysis of Eating Behavior Questionnaires

Obesity and overweight are well-documented risk factors for numerous diseases that negatively impact life expectancy and quality of life, including cardiovascular diseases, diabetes, and cancer.Although the effects of weight status on brain function have been extensively studied, the application of machine learning (ML) and deep learning (DL) techniques in this domain remains underexplored.This study aims to address this gap by creating a unique dataset comprising electroencephalography (EEG) data from 19 channels, recorded while participants with varying body mass indices were exposed to visual food cues.The primary objective was to classify the differences in brain signals between normalweight and overweight/obese individuals using advanced DL methods.To mitigate overfitting and data imbalance, tabular data augmentation was employed.Additionally, the Supervised Tabular Meta-Learning (SuperTML) method was utilized to embed EEG features into images, marking a novel application for this type of data.Classification results indicate that DenseNet-121 achieved the highest accuracy, with a rate of 0.97 at channel T4.Regionally, the temporal area yielded the best average accuracy rates.Furthermore, the study investigated the correlation between EEG data and eating behavior through regression analysis, applying Random Forest, eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Voting ensemble regression models to the participants' questionnaire responses.A significant relationship between EEG data and the questionnaires was identified, with the LightGBM regressor achieving an R value of 0.966.These findings demonstrate superior performance compared to existing literature in several aspects.This study underscores the potential of DL in enhancing our understanding of the neural mechanisms underlying eating behaviors in individuals with different body weights and provides a robust methodological framework for future research in this field.