Purpose <p>To develop regression models for predicting adolescent idiopathic scoliosis (AIS) progression using machine learning with clinical and radiographic data.</p> Methods <p>We included female patients aged 10 to 13 years diagnosed with AIS at our institution between 2007 and 2020. Only those with mild-to-moderate AIS (initial Cobb angle of 10°–40°) were analyzed. The objective variables were the Cobb angle at 1and 2 years later and at the last visit. A stacking regression metamodel with 5-fold cross-validation was developed using age, bracing, first menstruation, height, weight, Cobb angle at first visit, curve pattern, Risser sign, and kyphosis. Model performance was evaluated by mean absolute error (MAE) and R² with internal validation by 1000 bootstrap samples.</p> Results <p>We analyzed 201 patients. The median Cobb angle at the first visit was 27.0° (interquartile range, 21–32), and the median observation period was 44.0 months. The developed models achieved: MAE, 4.22 (standard deviation [SD]: 0.25), R², 0.73 (SD: 0.03) for the 1-year Cobb angle prediction; MAE, 5.24 (SD: 0.34), R², 0.67 (SD: 0.04) for the 2-year Cobb angle prediction; and MAE, 5.99 (SD: 0.35), R², 0.73 (SD: 0.03) for the prediction at the last visit.</p> Conclusion <p>The 1-year prediction outperformed previously reported models with shorter follow-up. Despite an extended observation period of nearly 4 years, prediction accuracy at the last follow-up remained clinically valuable, supporting the long-term reliability of the model.</p>