Purpose <p>This study aimed to develop a machine learning (ML) pipeline for predicting skin toxicity in breast cancer patients after radiation therapy. Specifically, we investigated whether integrating clinical-dosimetric parameters with shape radiomic features from computed tomography (CT) images could enhance prediction accuracy.</p> Method <p>We retrospectively analyzed data from 78 breast cancer patients. An experienced radiation oncologist manually delineated the Clinical Target Volume (CTV) and skin. A total of 12 shape radiomic features from the CTV and 32 clinical-dosimetric features were extracted. An extra tree classifier was employed for feature selection, and the selected features were subsequently used to train four ML models: Support Vector Machine (SVM), Random Forest (RF), Logistic Regression, and Gaussian Process Regression (GPR). Model performance was evaluated using accuracy, sensitivity, specificity, F1 score, and the area under the Receiver Operating Characteristic curve (ROC-AUC).</p> Result <p>ML models trained with the integrated feature set outperformed those based solely on clinical-dosimetric parameters. Notably, the ROC-AUC improved from 0.66 to 0.74. The RF model achieved the highest performance, with an AUC of 0.74, accuracy of 0.75, sensitivity of 0.74, specificity of 0.74, and an F1 score of 0.74.</p> Conclusion <p>Integrating shape radiomic features with clinical-dosimetric parameters significantly enhances the prediction of skin toxicity in breast cancer patients undergoing radiation therapy. These findings support the potential of the proposed ML pipeline to inform personalized treatment planning and mitigate skin toxicity risk in high-risk patients.</p>