Objective To explore the application value of a machine learning model based on a hierarchical topological architecture in personalized surgical planning for horizontal strabismus.Methods The clinical data of 608 patients who underwent horizontal strabismus surgery at Xiangya Hospital of Central South University from March 2021 to March 2024 were retrospectively analyzed. Fourteen routine preoperative indicators, including age, prism diopter, and axial length, were extracted, and an ensemble learning algorithm was used to construct a two-stage hierarchical topological model. The first stage of the model predicted the intervention choice among 8 surgical operations (recession or resection) corresponding to the 4 horizontal rectus muscles of both eyes, while the second stage predicted the specific surgical dose. The dataset was divided into a training set and a test set at a ratio of 80% to 20%. Model performance was evaluated using the macro-average area under the curve (macro-AUC), mean absolute error (MAE), root mean square error (RMSE), and the exact match rate of the surgical plans. The SHapley Additive exPlanations (SHAP) method was introduced for feature interpretation.Results There were no significant differences between the exotropia and esotropia groups in age, mean axial length, or mean spherical equivalent (P > 0.05). However, the prism diopter differed significantly between the two groups (P < 0.05). Preoperative prism diopter was positively correlated with the actual surgical doses of the major target muscle interventions, including left lateral rectus recession, right lateral rectus recession, left medial rectus resection, and right medial rectus resection (all P < 0.05). The hierarchical topological model achieved a macro-AUC of 0.962 for the surgical muscle selection task. For surgical dose prediction, the MAE and RMSE were 0.491 mm and 1.494 mm, respectively. The exact match rate for the complete surgical plan was 64.8% (79/122). Feature importance analysis showed that spherical equivalent and preoperative prism diopter were among the most important features driving the model's surgical planning decisions.Conclusions The hierarchical topological machine learning model can simulate the decision-making thinking of clinical experts with high precision, achieving highly precise personalized surgical planning, and exhibits strong potential for clinical decision support.