Determination of obesity tendency by processing bio-signals obtained from food stimuli

This thesis presents a unique dataset comprising Electroencephalography (EEG) data collected from 19 channels while participants with different body mass indices were exposed to visual food stimuli. Additionally, two different eating behaviour questionnaires were administered to these individuals. Classification and regression analyses were performed using the obtained data. The primary aim is to classify the differences in brain signals between normal weight and overweight/obese individuals using machine learning and deep learning methods. In the first part of the study, classification is performed using traditional machine learning methods, achieving high classification accuracy. In the second part, deep learning methods are employed. Tabular data augmentation is used to reduce overfitting and data imbalance. Furthermore, the Supervised Tabular Meta Learning (SuperTML) method is utilized to embed EEG features into images, marking a novel application for this type of data. The classification results indicate that DenseNet-121 achieved the highest accuracy with a rate of 0.97 in the T4 channel. Regionally, the temporal area provided the best average accuracy rates. The study also investigates the correlation between EEG data and eating behaviour through regression analysis by applying Random Forest, eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Voting regressor models to the participants' questionnaire responses. A significant relationship between EEG data and questionnaires is identified, with the LightGBM regressor reaching an R² value of 0.966. These findings demonstrate superior performance in many aspects compared to the existing literature. This study underscores the potential of deep learning and machine learning to enhance our understanding of the neural mechanisms underlying eating behaviours in individuals with different body weights and provides a robust methodological framework for future research in this area.