Can Machine Learning Algorithms Classify Inner Speech from EEG Brain Signals?

For the severely disabled who still possess a normal level of consciousness, currently, available assistive devices are mostly incompatible, making the development of an inner speech-controlled Brain-Computer Interface (BCI) highly sought after. Using electroencephalograms, brain signals associated with inner speech can be recorded and subsequently classified using a machine learning algorithm, allowing for potential control over an assistive BCI device. Using an inner speech dataset consisting of four directional cues, this paper reports that a developed random forest classification model possessed an accuracy of 29.7%, an SVM classification model possessed an accuracy of 35.3%, and a KNN classification model possessed an accuracy of 31.4% when classifying between "up," "down," "left," and "right" inner speech. Although the accuracy of the classification model used in this study is low, the results provide promising evidence that machine learning algorithms can be used to create more complex BCI devices with further research and progress. It also demonstrates the potential to develop internal voice-controlled BCI devices using machine learning algorithms, opening up a new field of assistive technological innovation.