Hypothesis: Electroencephalography and Artificial Neural Networks can be combined to read in a user’s EEG-based brain activity and then to correctly classify that activity.
Goal: This research project aims to combine EEG (electroencephalography) and ANNs (artificial neural networks) by reading in a user's EEG-based brain activity and using an ANN to correctly classify that activity. This specific application aims to classify EEG data of a user being presented with digits (0-9) and letters (A-J).
Process: The project goals are accomplished by building an EEG headset capable of collecting data, generating a labeled dataset (EEG activity is the data, character being presented is the label), and creating ANNs to analyze the labeled dataset.
Results: The EEG headset based on the UltraCortex III from OpenBCI was successfully built. A data collection protocol was created, programs were coded to facilitate this data collection, and the dataset was successfully generated (3160 samples total, 158 samples/character). Several ANNs were created, and these networks were capable of learning and of overfitting on the data, but the classification on test data did not reach accuracy levels beyond chance (more time needs to be devoted in trying different networks and manipulating the data).
Future work: Tasks which are recommended for continuing this project include adjusting network parameters of existing CNNs, trying a wider variety of neural network architectures, trying data mining techniques, extracting more features in different ways from the existing data, collecting more digit and letter EEG data, altering data collection process.