A Comparison of Machine Learning Models for Classification of Parkinson’s Disease During a Working Memory and Sustained Attention Task

Description

Mercedes A. Terry and six co-authors compare machine learning models that sort people with Parkinson disease from healthy controls using brain and eye signals recorded during thinking tasks, in 2026 work. Participants performed a working memory task and a sustained attention task while scalp electrodes tracked brain waves and a camera tracked pupil size. Behavior alone often fails to expose early thinking change in this disease, so the team aligned signal features to task moments.

The combined signal pipeline separated the groups more accurately than standard behavior scores did. Seven features carried most of the sorting power, including frontal brain rhythm markers, rhythm ratios, and pupil responses. The telling features clustered in the hardest task stretches, with heavier memory loads, longer waits, and wrong answers. The authors offer the setup as a readable platform for hunting early markers, pending larger proof studies.

Catalog record

Source: https://europepmc.org/article/MED/PMC13510374