Decoding Scalp EEG Signatures of Attentional Switching Using Machine Learning

Poster Session E - Monday, March 9, 2026, 2:30 – 4:30 pm PDT, Fairview/Kitsilano Ballrooms

Melih Keskin¹, Simrit Rai¹, Sairamya Nanjappan Jothiraj¹, Julia Kam¹; ¹University of Calgary

The ability to switch between external and internal attention is critical for everyday life. Despite its prevalence and wide-ranging impact on cognitive functions, the value of using electrophysiological features to decode these types of attentional switches remains underexplored. To address this, we recorded scalp EEG from 37 participants while they performed an attentional switching task requiring them to switch between attending externally to tones or internally to their thoughts. We then used machine learning models to detect the direction of these switches (i.e., switching to external versus switching to internal attention). For these analyses, we extracted block-level spectral features across four canonical frequency bands (theta, alpha, beta and gamma) from 63 electrodes using the means and the standard deviations of brain activity across the three second period of switching, resulting in a total of 504 features for each participant. To enhance classification performance, we selected the fifteen most informative features using the random forest algorithm across participants. These features served as inputs into several classification models. The model with superior performance was the 5-fold within-subject support vector machine classification model, which achieved a Matthew’s Correlation Coefficient of 0.62. The most informative EEG features consisted of beta and gamma band activity over prefrontal, frontocentral, and centro-parietal regions, which consistently emerged across multiple iterations of feature selection. These results suggest EEG features can be used to predict switching between external and internal attention at above chance levels, highlighting their potential value for real-time prediction in future studies.

Topic Area: ATTENTION: Other