Amyloid beta (Aβ) disease-modifying therapies (DMT) highlight the need for biologically informed diagnoses of Alzheimer’s disease (AD). Early-stage biomarkers, such as fluid biomarkers and amyloid PET in asymptomatic individuals, present uncertainties regarding disease progression and intervention timing. Effective tools for predicting progression in individuals with early Aβ changes are crucial for personalized treatment. This study developed machine learning (ML) models to predict Aβ plaque burden over five years in subjects with low initial plaque levels (Centiloid < 24). Predictive features included demographics, APOE genotype, cognitive tests, regional PET SUVRs, and volumes. Three ML classification models—support vector machines (SVM), random forests (RF), and multilayer perceptions (MLP)—were trained and cross-validated on ADNI and independently tested on OASIS-3. Feature contributions were analyzed using Shapley Additive Explanations (SHAP). The models demonstrated strong predictive performance, with weighted F1 scores of 0.86 (SVM), 0.85 (MLP), and 0.87 (RF) on ADNI, and 0.82 (SVM), 0.83 (MLP), and 0.80 (RF) on OASIS-3. SHAP analysis identified the prefrontal region as the most significant predictor, aligning with prior research and enhancing our understanding of amyloid plaque dynamics. These findings support the use of ML-based tools in improving early AD diagnosis and optimizing intervention planning.