Development of next generation sentence classification models using EEG signals

Electroencephalography (EEG) signals are described as the letters of the brain, and computer scientists aim to obtain meaningful sentences by combining these letters. For this reason, processing EEG signals is a hot topic for neuroscience and machine learning. However, EEG processing and classification studies in the literature are generally related to disease detection and emotion detection. Within the scope of this thesis, a new project has been started to extract more information from EEG signals and the subject of this project is EEG sentence classification. In this thesis study, two EEG sentence datasets were collected from 40 volunteer participants, 20 participants for each dataset. The collected data sets contain 20 classes and each of these classes represents a sentence. Micro descriptors and graph-based feature extractors are used to propose automatic classification models. The proposed square-sum graph pattern-based EEG signal classification model achieved a classification accuracy of 99.19%. In this study, the feature extraction capability of a function based on the sum-square graph was investigated. The DSBP-IMCMV based model, on the other hand, achieved the best overall classification rates with 98.81% and 98.19%, respectively, in showing and listening modes. The results clearly showed that sentence classification can be done with EEG signals, and this thesis is one of the first theses in the field of EEG sentence recognition/classification. The results obtained clearly show the success of the thesis study.