DEF-DSVM: A deep ensemble feature learning and deepSVM approach for multifaceted analysis and diagnosis of Alzheimer’s disease from EEG signals

• Published in: Methods (San Diego, Calif.) Vol. 242; pp. 169 - 186
• Main Authors: Hesari, Shabnam, Ghaffari, Hamidreza, Rezaee, Khosro
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 01.10.2025
• Subjects: Aged; Alzheimer Disease - diagnosis; Alzheimer Disease - physiopathology; Alzheimer’s disease; Cognitive Dysfunction - diagnosis; Cognitive Dysfunction - physiopathology; Deep ensemble feature learning; Deep Learning; Deep support vector machine; Diagnosis, Computer-Assisted - methods; Early Diagnosis; Electroencephalography; Electroencephalography - methods; Female; Humans; Male; Mild cognitive impairment; Signal Processing, Computer-Assisted; Support Vector Machine; Wavelet Analysis; Deep support vector machine; Early diagnosis; Electroencephalography; Alzheimer’s disease; Mild cognitive impairment; Deep ensemble feature learning
• ISSN: 1046-2023; 1095-9130; 1095-9130
• DOI: 10.1016/j.ymeth.2025.08.003

•DEF-DSVM achieves 98.17% accuracy in early Alzheimer’s diagnosis using EEG data.•EEG wavelet and spectrogram analysis reveal key insights into Alzheimer’s and MCI.•Leave-One-Subject-Out validation ensures robust generalization on real EEG data.•DeepSVM combines deep learning features with SVM precision for diagnostics. Early detection of Alzheimer’s disease (AD) and its precursor, mild cognitive impairment (MCI), is paramount for timely intervention and effective disease management. This study introduces a novel computer-aided diagnostic model that leverages electroencephalogram (EEG) data to precisely identify and classify AD and MCI. A comprehensive preprocessing pipeline is employed, incorporating discrete wavelet transform (DWT) for EEG signal decomposition into relevant subbands and subsequent signal windowing to address non-stationarity. Spectrograms derived from these preprocessed signals serve as input for a deep ensemble feature learning and deep support vector machine (DEF-DSVM) architecture. The DEF-DSVM model significantly enhances the accuracy of diagnosing both MCI and AD, achieving an impressive 98.17% accuracy rate that surpasses contemporary state-of-the-art methods. Beyond diagnostic precision, the model effectively identifies specific EEG subbands—namely alpha, theta, and delta—instrumental in elucidating the pathophysiology of AD and MCI. The structure’s generalizability and robustness are validated using the Figshare dataset, encompassing, AD, MCI, and control classes. To ensure a rigorous assessment of the model’s performance, the Leave-One-Subject-Out (LOSO) cross-validation procedure is employed in lieu of the traditional K-fold approach, mitigating the risk of overoptimistic performance estimates and providing a more accurate reflection of the model’s ability to generalize to novel, unseen subjects. Further evaluation of the method’s generalizability through its application to an EEG dataset related to attention deficit hyperactivity disorder (ADHD) highlights its broader clinical utility across various neurodegenerative disorders. These findings establish the DEF-DSVM model as a reliable and potent tool for the early diagnosis and monitoring of AD and MCI, offering substantial accuracy gains and demonstrating its potential for widespread application across different neurological conditions.