<h3>Purpose/Objective(s)</h3> Radiation-induced dermatitis (RD) is one of the major toxicities in patients with breast cancer (BC) receiving adjuvant radiation treatment (RT), which significantly affects the patient's treatment tolerance and quality of life and potentially compromises tumor control probability. However, to differentiate the degree of RD could be challenging due to the wide variations related to current methods used to evaluate the skin toxicity symptoms. In this study, we developed, for the first time to our knowledge, a machine learning framework synergistically integrating multi-region dose gradient correlated-radiomics features extracted from pre-treatment planning CT images and dosimetric factors with clinical variables for early prediction of RD. <h3>Materials/Methods</h3> A total of 214 patients with BC receiving RT between 2019 and 2021 from four cancer centers were included in this study. Three machine learning classifiers (a 5-fold cross-validation random forest, Radial Kernel svm, and gradient boosting decision tree (GBDT)) were employed for model training and validation. Radiomics features extracted from 3 PTV ROIs and 3 skin dose-related ROIs on planning CT image and 14 dosimetric factors as well as 17 clinical variables were used to train machine learning models. The utility of the constructed models in predicting RD was evaluated by the ROC analysis in the validation cohort. <h3>Results</h3> The random forest model combining 10 radiomic features, 3 dosimetric factors and 6 clinical variables achieved a better performance with an AUC value of 0.946 (95%[CI]: 0.887-0.987) compared with GBDT (AUC=0.912, 95%[CI]: 0.839-0.985) and SVM model (AUC=0.844, 95%[CI]: 0.786-0.913). The most valuable influence factors for model performance include 10 radiomics features (PTV105.F8.ShapeMeanBreadth, PTV108.F1.GOH0.975Quantile, PTV108.F2._GLCM25180.1Dissimilarity, SKIN20.F2._GLCM25225.4Contrast, SKIN20.F4.ID_GlobalMin, SKIN30.F2._GLCM25225.4Contrast, SKIN30.F6.IHGaussFit1GaussStd, SKIN30.F8.ShapeMax3Ddiameter), 3 dosimetric factors (PR, EQD2_all, fractionation.regimen.Gy.fx.), and 6 clinical variables (laterality quadrant positions, histologic type, T stage, hormone therapy and lotion application). <h3>Conclusion</h3> The machine learning models depicting the correlation of dose gradient related radiomics features, dosimetric factors and clinical variables can be used to accurately predict the severe RD in patients with BC receiving RT, which can potentially improve the effectiveness of RT for BC from a precision treatment perspective.