Development of intelligent methods for emotion recognition from EEG signals

Nowadays, emotion recognition has become an important element in many areas.EEG signals are thought to diagnose a variety of brain and neurological conditions. EEG signals were used in this study because they generate characteristic signals for the detection and analysis of the emotional state. In the proposed model of this thesis, it was aimed to develop an effective and simple method to develop high-accuracy automatic emotion recognition,together with developing a higher classification with a new learning model for one-dimensional signals, and to achieve high-accuracy automatic emotion recognition. In the first proposed methods, a new automatic EEG emotion recognition model is presented using local binary pattern, multilevel discrete wavelet transform, neighborhood component analysis and k-nearest neighbor classifier. This model has reached an excellent classification rate of 100.0% in the GAMEEMO dataset. These results clearly demonstrated the model's high classification ability on EEG signals for emotion classification. In the other proposed method, the prime pattern and tunable q-factor wavelet transform (TQDD) feature generation model are presented. This model covers all phases of the machine learning model, including feature extraction, feature selection, and classification. Since the proposed model generates 87 feature vectors, this model is named PrimePatNet87. The publicly available GAMEEMO, DREAMER and DEAP datasets were used to develop the proposed model. The PrimePatNet87 model achieved over 99% classification accuracy across all datasets with leave one subject out (LOSO) validation. These results show that the proposed prime pattern model is ready for real-world applications.