Electroencephalograms (EEG) and magnetoencephalography (MEG) are techniques for measuring, directly or indirectly, actual or relative changes in voltage throughout the body. This type of time series data is often related to classification problems. EEG and MEG are also central to human computer interface research. The question we wish to address on this project is, what is the best way to make predictions from ECG/EEG and MEG? We will treat the problem as time series classification. We have an existing database of publicly available labelled EEG/MEG datasets, tools in both Python for time series classification and collaborators who are experts in traditional techniques for analysing EEG/MEG. Our ultimate goal is an effective automated end-to-end pipeline for EEG/MEG classification using open source software such as sktime. This would encourage reproducible research in the field and facilitate more widespread analysis. This ten-week project will contribute to this goal and will involve three phases.
The first will involve implementation of a pipeline benchmark approach of feature extraction and standard classification. This will require standard toolkits for features such as wavelets as soft dependencies in sktime and the evaluation of existing techniques with a range of transformation/scikit-learn classifier combinations.
The second phase will involve the development of existing time series classification algorithms to work directly with EEG data.
The final phase will be to help construct and manage an experimental scheme to compare the performance of a range of algorithms on problems of this kind.
Mentor: Tony Bagnall
