Statistical Learning for Personalized Prediction of Alzheimer's Disease Progression: A Survey of Methods, Data Challenges, and Future Directions

Alzheimer's disease (AD) develops and progresses uniquely in each individual, making personalized predictions critical for early intervention and effective treatment. Advances in statistical learning and artificial intelligence have shown promise in the analysis of large, multimodal, and longitudinal AD data to generate meaningful insights for individualized care. Current methods can be largely categorized into four main areas: (i) probabilistic state-space models that track latent disease stages while providing uncertainty estimates; (ii) deep neural networks, including convolutional, recurrent, and transformer architectures, that uncover complex spatial and temporal dynamics; (iii) graph neural networks that capture brain connectivity structures; and (iv) digital twins for simulating individualized disease trajectories. Key challenges, such as high-dimensional features, missing data, class imbalance, and computational demands, are being tackled through approaches such as Bayesian shrinkage, low-rank tensor factorization, cross-site harmonization, and generative data augmentation. The field is evolving toward multimodal integration, interpretable and uncertainty-aware predictions, and secure federated learning. Ongoing needs include hybrid causal deep learning models, scalable optimization techniques, and ethical deployment strategies to accelerate the development of clinically viable tools for personalized prognoses and interventions of AD. This article is categorized under: Statistical Learning and Exploratory Methods of the Data Sciences > Modeling Methods Statistical and Graphical Methods of Data Analysis > Modeling Methods and Algorithms Data: Types and Structure > Image and Spatial Data.