Machine learning-based categorical emotion classification model using EEG signals

In recent years, automatic emotion detection and classification is one of the most frequently studied topics in the literature. Emotions are effective in individuals' relations with the outside world, in their decisions and actions. Therefore, emotion recognition has an important role in human-computer interaction. Literature studies show that EEG signals can detect some neurological and cerebral activities in detecting emotions. In this thesis, EEG signals were used as distinctive signals were produced for the detection and analysis of emotions. In this context, it is aimed to develop a new machine learning model for automatic emotion recognition with high accuracy by creating a unique data set and proposing effective methods for EEG signal interpretation. With this newly developed model, all stages of the machine learning model, including feature extraction, feature selection and classification, were carried out using the original data set. In this study, a feature extractor based on VitC-Pat (Vitamin C Pattern) is proposed. In addition, the Discrete Wavelet Transform (DWT) based decomposition method was used to separate the EEG signals into subbands. In the feature extraction phase of the developed model, VitC-Pat, Local Binary Patterns (LBP) and statistical methods were used. In the second phase of the machine learning model, feature selection is applied. At this stage, the Iterative Neighborhood Component Analysis (INCA) method was used, and in the last stage of the developed model, the classification process was carried out using Support Vector Machines (SVM). With this model, which was tested on the data set originally collected within the scope of the thesis study, accuracy values of 92.74% for the Arousal axis and 96.85% for Valence were provided.