Application of machine learning methods for the analysis of functional neuroimaging signals, or 'brain-function decoding', is a highly interesting approach for better understanding of human brain functions. Recently, Kauppi et al. presented a brain-function decoder based on a novel feature extraction approach using spectral LDA, which allows both high classification accuracy (the authors used sparse logistic regression) and novel neuroscientific interpretation of the MEG signals. In this thesis we evaluate the performance of their brain-function decoder with additional classification and input feature scaling methods, providing possible additional options for their spectrospatial decoding toolbox SpeDeBox. We find the performance of their brain-function decoder to validate the potential of high frequency rhythmic neural activity analysis, and find that the logistic regression classifier provides the highest classification accuracy when compared to the other methods. We did not find additional benefits in applying prior input feature scaling or reduction methods.
