The study decided to use machine learning models for predicting** hematologic toxicity (HT), one of the most severe side effects of chemoradiotherapy in the advanced stages of cervical cancer** patients. Two imaging data types were taken into consideration: **radiomics** (features from CT images) and **dosiomics** (features from radiation dose distributions). The study conducted a data analysis of 205 patients. After extracting hundreds of features and performing feature selection, three models were built **white** being radiomics-only, dosiomics-only, and a hybrid model combining both. XGBoost was used for all models, and each one was evaluated in terms of sensitivity, specificity, and AUC. The **hybrid model was the most successful**, reaching **AUC=0.83**, and outdoing the single-source models. After that, SHAP (SHapley Additive exPlanations) was applied to the model to make it interpretable and to identify the features that affected the predictions and their significance. In the end, the combination of radiomics and dosiomics not only increased prediction accuracy but also contributed to transparency of the model.