Background <p>This study aimed to develop and validate a data-driven framework using a radial basis function (RBF) neural network to predict site-specific alveolar bone remodeling during mandibular incisor decompensation in patients with skeletal Class III malocclusion.</p> Methods <p>This retrospective cohort study included 10 patients (39 mandibular incisors). Pre- and post-treatment cone-beam computed tomography scans were registered in three dimensions. A novel geometric framework quantified tooth movement using four parameters: rotation radius, chord length, rotation angle, and vertical displacement. An RBF neural network was trained to model the nonlinear relationship among pre-orthodontic morphology, tooth movement patterns, and post-treatment alveolar bone thickness. Model performance was evaluated using the mean squared error and coefficient of determination (R²).</p> Results <p>Distinct alveolar bone response patterns were identified. Compression-side resorption approached a biological limit near 0&#xa0;mm. Tension-side apposition exhibited spatially graded efficiency, increasing from approximately 0% at the cervical level to 50–60% at the apical level. The RBF-based predictive model achieved high accuracy (R² = 0.7–0.9) and effectively captured the nonlinear interplay between movement types and local anatomical constraints.</p> Conclusions <p>The established framework accurately quantifies and predicts site-specific alveolar bone responses. Bone remodeling is governed by the synergistic interaction between tooth movement patterns and intrinsic anatomical limitations. This framework enables preemptive risk assessment and supports personalized orthodontic treatment planning.</p>