Location
Online
Event Website
https://hicss.hawaii.edu/
Start Date
3-1-2023 12:00 AM
End Date
7-1-2023 12:00 AM
Description
The classification of sensorimotor rhythms in electroencephalography signals can enable paralyzed individuals, for example, to make yes/no decisions. In practice, these approaches are hard to implement due to the variability of electroencephalography signals between and within subjects. Therefore, we report a novel and fast machine learning model, meeting the need for efficiency and reliability as well as low calibration and training time. Our model extracts finely graded frequency bands from motor imagery electroencephalography data by using power spectral density and training a random forest algorithm for classification. The goal was to create a non-invasive generalizable method by training the algorithm with subject-independent EEG data. We evaluate our approach using one of the currently largest publicly available electroencephalography datasets. With a balanced accuracy of 73.94%, our novel algorithm outperforms other state-of-the-art non-subject-dependent algorithms.
Recommended Citation
Penava, Pascal; Brozat, Marie-Louise; Zimmermann, Yara; Breitenbach, Johannes; Ulrich, Patrick; and Buettner, Ricardo, "Subject-Independent Detection of Yes/No Decisions Using EEG Recordings During Motor Imagery Tasks: A Novel Machine-Learning Approach with Fine-Graded EEG Spectrum" (2023). Hawaii International Conference on System Sciences 2023 (HICSS-56). 3.
https://aisel.aisnet.org/hicss-56/in/hci/3
Subject-Independent Detection of Yes/No Decisions Using EEG Recordings During Motor Imagery Tasks: A Novel Machine-Learning Approach with Fine-Graded EEG Spectrum
Online
The classification of sensorimotor rhythms in electroencephalography signals can enable paralyzed individuals, for example, to make yes/no decisions. In practice, these approaches are hard to implement due to the variability of electroencephalography signals between and within subjects. Therefore, we report a novel and fast machine learning model, meeting the need for efficiency and reliability as well as low calibration and training time. Our model extracts finely graded frequency bands from motor imagery electroencephalography data by using power spectral density and training a random forest algorithm for classification. The goal was to create a non-invasive generalizable method by training the algorithm with subject-independent EEG data. We evaluate our approach using one of the currently largest publicly available electroencephalography datasets. With a balanced accuracy of 73.94%, our novel algorithm outperforms other state-of-the-art non-subject-dependent algorithms.
https://aisel.aisnet.org/hicss-56/in/hci/3