Document Type : Original/Review Paper

Authors

• Amir Hossein Firoozjah ¹
• Ali Mehri ²
• Maryam Jamaati ³

¹ Department of Physics, Babol Noshirvani University of Technology

² Associate Professor in Physics Department of Babol Noshirvani University of Technology

³ Faculty of Computer Engineering, Iranian eUniversity, Tehran, Iran

Abstract

Alzheimer’s disease (AD) is a prevalent, costly, and fatal neurodegenerative disorder that impairs quality of life. Machine learning shows promise for analyzing healthcare data to aid early diagnosis. This study evaluates 21 machine learning algorithms (supervised, deep, ensemble, and hybrid) trained on four frequency-domain and three time-domain EEG features from 65 adults (36 with AD, 29 without) using an OpenNeuro dataset. Results showed that deep learning algorithms trained on frequency-domain features outperformed those using time-domain features. Specifically, a convolutional neural network (CNN) with wavelet transform feature achieved the highest accuracy of 97.05%. Unlike previous studies that generally evaluated a limited number of algorithms from a single learning paradigm or employed heterogeneous feature extraction pipelines, the present study provides a comprehensive and unified comparative benchmark of 21 representative algorithms spanning four learning families under an identical experimental framework. These findings demonstrate the comparative effectiveness of different machine learning paradigms for EEG-based Alzheimer's disease classification and highlight the promising role of deep learning within the adopted benchmarking framework.

Keywords

• Alzheimer’s Disease
• Electroencephalography Data
• Machine Learning
• Deep Learning
• Ensemble Learning

Main Subjects

• H.6.5.13. Signal processing