21th AIAI 2025, 26 - 29 June 2025, Limassol, Cyprus

Classifying cognitive states of Alzheimer’s Disease with machine learning using digital biomarkers from the Bio-Hermes study cohort

Eoin Houstoun, Daniel Stamate, Henry Musto, David Reeves, Catharine Morgan, Roxana Hutanu, Kalliopi Mavromati, Dorina Cadar, Daniel Stahl

Abstract:

  Alzheimer’s Dementia (AD) is a progressive neurodegenerative disorder with clinical stages ranging from cognitively normal (CN) to mild cognitive impairment (MCI) and mild Alzheimer’s Dementia (AD). Accurate classification of these stages is critical for early intervention and treatment planning, and AD risk prediction is also important for patient stratification. This work explores the predictive capabilities of machine learning models using a combination of digital biomarkers from the Bio-Hermes study. Five models—Logistic Regression, Elastic Net, Random Forest, Gradient Boosting, and XGBoost—were evaluated on a 20% test set after being trained or hyperparameter-tuned using repeated cross-validation, on an 80% training set. XGBoost achieved the best overall performance with an AUC of 0.815 followed closely by Gradient Boosting (AUC = 0.806). The findings highlight the potential of ensemble methods like XGBoost for the prediction of Alzheimer’s clinical stages and risk of AD, with digital biomarkers proving to be valuable while non-invasive and inexpensive to produce predictors. This is an important aspect in the economy of a very expensive and challenging to diagnose, treat, and evaluate risk for, medical condition such as Dementia.  

*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.