Adaptive learning based on cognitive load using artificial intelligence and electroencephalography
Master thesis
Permanent lenke
http://hdl.handle.net/11250/2457138Utgivelsesdato
2017Metadata
Vis full innførselSamlinger
Sammendrag
The domains of artificial intelligence and cognitive sciences have not been properly explored within instructional design and learning systems, despite being optimistic in their positive effects when utilized correctly in said systems. Passive measurements of the cerebral cortex allows for accurate signal extraction, which is why EEG-equipment has become popular in brain-computer interfaces, as well as the base for feature extraction in machine learning mechanisms within these. A literature study has been performed to gain insights into good practice and past experiences in utilizing both AI and EEG-equipment in classification tasks. The relative low cost of the Emotiv EEG-headsets enables more experiments to be conducted in many applications and domains, which is beneficial for proof-of-concepts and exploring the feasibility of their application. In this project a driver has been utilized to sample raw sensor-data from the Emotiv Epoc EEG-headset, and a recurrent neural network has been constructed to classify fifteen different emotions from raw data provided by Swartz Center for Computational Neuroscience. The RNN has shown great performance in emotion classification from multi-channel EEG-signals, achieving 99\% accuracy for both training- and test-sets. The RNN did not generalize enough for practical usage in real-time sampling using the Emotiv Epoc EEG-headset on a new subject, but its qualities can possibly be utilized in further work by re-training it on a customized data set.