Robust Multi-Modal Machine Learning Model for Early Detection and Progression Prediction of Alzheimer's Disease Using Eeg-Based Signal Dynamics

Abstract

Alzheimer's disease (AD) is the most prevalent neurodegenerative disorder and the leading cause of dementia 
worldwide, affecting more than 55 million individuals and imposing substantial social and economic costs. Early 
diagnosis of AD and accurate identification of Mild Cognitive Impairment (MCI), a transitional stage between normal 
aging and dementia, are crucial for enabling timely intervention and slowing disease progression. 
Electroencephalography (EEG) has emerged as a promising diagnostic tool due to its non-invasive nature, low cost, 
and ability to capture neural activity with high temporal resolution. However, many existing EEG-based diagnostic 
systems rely on limited feature sets and single-model architectures, restricting their ability to fully exploit the complex 
patterns associated with Alzheimer's pathology. To address these limitations, this study proposes the Multi-Modal 
Machine Learning Alzheimer's Detection (MM-MLAD) framework, which integrates multi-domain EEG features, 
neuropsychological assessment scores, and demographic information within a unified predictive architecture. The 
framework extracts comprehensive EEG biomarkers from time, frequency, time-frequency, and functional connectivity 
domains and combines them with cognitive assessment measures such as MMSE, CDR, and MoCA. Multiple learning 
models, including 1D-CNN, Bidirectional LSTM, Transformer, and XGBoost, are trained in parallel and fused through 
a stacking ensemble strategy to enhance diagnostic performance. Experimental evaluation on a dataset of 366 subjects 
demonstrates that MM-MLAD achieves 94.7% classification accuracy, a weighted F1-score of 94.4%, and an AUC
ROC of 0.981, while also providing reliable prediction of MCI-to-AD progression. Furthermore, explainable AI 
techniques reveal that theta-band slowing, reduced alpha coherence, and elevated delta activity are the most influential 
biomarkers, highlighting the framework's clinical relevance and potential for supporting early Alzheimer's diagnosis 
and prognosis.