EEG-Based Alzheimer's Disease Classification — Multi-Center Study (Code)

EEG-Based Alzheimer's Disease Classification — Multi-Center Study (Code) This repository contains the code accompanying the paper on EEG-based classification of Alzheimer's disease (AD), Mild Cognitive Impairment (MCI), Frontotemporal Dementia (FTD) and Cognitively Normal (CN) controls across six centers in Chile, Argentina, Colombia, Brazil and Turkey. The repository is organized into two parts: pipeline/ — A machine learning pipeline that reads EEG markers and trains/evaluates Random Forest classifiers under three validation strategies: within-center (StratifiedGroupKFold), multi-center (StratifiedGroupKFold across all centers), and cross-center (LeaveOneGroupOut). Supports repeated cross-validation (500 repeats), probability calibration, and SHAP feature importance analysis. analyses/ — Jupyter notebooks that reproduce all figures and statistics reported in the paper, including classifier performance (AUC, precision, sensitivity), permutation-based p-values, calibration curves, SHAP beeswarm plots, feature stability, cross-strategy correlations, and correlations with neuropsychological test scores. Data availability: The input data required to run this code is available at: Belloli, L. (2026). EEG and neuropsicological tests dataset used for the multicentric generalization of EEG classifiers for dementia [Data set]. Zenodo. https://doi.org/10.5281/zenodo.20206227 Setup instructions and full documentation are provided in the README.