Automated artifact detection in infant EEG (Robin Kemmerich, University of Bremen)

Automated artifact detection in infant EEG: a comparison of machine and deep learning classifiers Infant EEG is highly artifact-prone, and manual artifact detection is labor-intensive and subjective. We evaluated three classifiers, Random Forest (RF), Support Vector Machine (SVM), and a deep learning (DL) model, for automated artifact detection on filtered infant EEG without feature extraction. Data came from 294 infants (mean age 8.34 months) in the BRISE cohort, yielding 66,851 expert-annotated epochs, of which 45% were artifacts. RF and DL achieved high balanced accuracy (.873 and .881), clearly outperforming SVM (.756). Higher classifier certainty improved accuracy but reduced data coverage. RF performed better on small training sets, DL on larger ones. Automated detection can standardize preprocessing and improve reproducibility in developmental research.