Detecting emotions using deep learning and wearable sensors

This thesis investigates applying deep learning techniques for classifying Electroencephalography (EEG) signals related to aesthetic experiences. EEG-based emotion recognition has been widely explored, but little attention has been given to aesthetic experiences involving complex neural dynamics. This research examines the use of Convolutional Neural Networks (CNN), Fully Connected Neural Networks (FCNN), Long Short-Term Memory (LSTM) networks, and Shallow FCNN to classify EEG signals from participants viewing art in a naturalistic gallery setting. Three classification tasks are examined: distinguishing Aesthetic signals from Baseline and Non-Aesthetic signals (Aes_vs_baseline_noAes), identifying Art-related vs. Baseline moments (Art_vs_baseline), and classifying Aesthetic vs. Non-Aesthetic moments (Aes_vs_noAes). The models are evaluated based on accuracy, precision, recall, and F-measure with and without feature selection methods like Principal Component Analysis (PCA) and Minimum Redundancy Maximum Relevance (mRMR). The results show that CNN consistently outperforms other models across the classification tasks, particularly in the Art_vs_baseline and Aes_vs_baseline_noAes tasks. Feature selection methods, however, did not significantly improve model performance. This study demonstrates the potential of deep learning for understanding neural processes related to aesthetic experiences and suggests that CNN is the most reliable approach for this task.