A comparison of classification algorithms in session-to-session generalisation in brain pattern recognition
Independent thesis Basic level (degree of Bachelor), 10 credits / 15 HE creditsStudent thesis
A brain computer interface (BCI) is a system to interpret a user’s intentionthrough using the users brain activity. The system is todayforemost used for research for disabled people. Electroencephalography(EEG) is the most common method used by BCI to record the brainactivity of the user, where the relevant information is read from thebeta and mu waves in the brain. These brain waves are from the partof the brain that is activated when user are doing problem solving orimagining moving a part of the body.This report will investigate EEG-data taken from test subjects imaginingto move left or right arm. To investigate the data from the EEG,two of the most common classification models, Multi-Layer Perceptron(MLP) and Support Vector Machines (SVM), will be used. The investigationwill show how well the models performs in session-to-sessiongeneralisation, which means we will evaluate the performance of theclassifier when testing the it on data from another session than the sessionit was trained on.The tests shows that the SVM, with the Gaussian Radial BasisFunction (RBF) as kernel, yielded the best results. For every personthere is three sessions, EEG data required at three separate occasions,that were used in the tests and, for the SVM, it is clear that trainingon the first two sessions combined and testing on the last session gavethe best classification of the EEG data. However, for the MLP thedifferences in results, from training on one session or two, smaller.
Place, publisher, year, edition, pages
2014. , 34 p.
Computer Engineering Computer Science
IdentifiersURN: urn:nbn:se:kth:diva-145969OAI: oai:DiVA.org:kth-145969DiVA: diva2:721285
Subject / course
Bachelor of Science in Engineering - Computer Engineering
Herman, Pawel, Dr
Ekeberg, Örjan, Prof.