In-Silico Prediction of Alzheimer’s Disease Using Computational Approaches

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder in which biological changes may appear prior to cognitive decline. The increasing availability of clinical, cognitive, neuroimaging, genetic, and fluid biomarker data has made machine learning (ML) a potentially suitable approach for identifying patterns associated with AD diagnosis and future progression. This review focuses on how the choice of data, prediction task, and validation strategy influences the reported performance of certain ML models in AD research. Conventional approaches, including logistic regression, support vector machines, and ensemble methods, remain useful for structured or engineered features, whereas deep learning models, particularly convolutional neural networks, are widely used in neuroimaging. Evidence from systematic reviews and cross-cohort studies suggests that high performance is obtainable in distinguishing established AD from cognitively unimpaired individuals, but predicting progression from mild cognitive impairment (MCI) is comparatively complex. Direct comparison of reported accuracy or area under the receiver operating characteristic curve (AUC) becomes difficult due to differences in cohorts, preprocessing, feature selection, data splitting, and external validation. Recent methodological evidence also shows that data leakage and inadequate control of confounding can substantially inflate reported performance. Therefore, the central challenge for ML is no longer demonstrating that prediction is possible, but establishing whether predictions remain reliable across independent populations and clinically relevant settings. Future research should prioritise subject-level validation, external and prospective testing, multimodal but clinically feasible data integration, interpretable modeling and transparent reporting. These considerations are essential if ML is to move from promising retrospective research applications toward reproducible clinical decision support.