Breast cancer remains a leading cause of morbidity and mortality among women worldwide. Existing prediction models relying solely on either psychosocial risk factors (e.g., age, family history, reproductive history) or clinical imaging/biopsy data face important limitations. Psychosocial models are applicable for population-level screening but yield low to moderate predictive power (AUC ~ 0.60), while clinical models (e.g., those based on Wisconsin Breast Cancer Dataset - WBCD) demonstrate high accuracy (AUC > 0.95) but are only applicable post-lesion detection. This paper proposes a novel multimodal machine learning approach that integrates psychosocial data collected through the nationally recognized Iskra survey and synthetic clinical features inspired by WBCD. The integrated dataset is processed through a pipeline that includes imputation, standardization, encoding, and class balancing using SMOTE. A stacking ensemble classifier combining Random Forest and XGBoost is developed and evaluated. Results show that the integrated model achieves superior predictive performance (AUC ~ 0.83) compared to unimodal models, while maintaining robustness in scenarios with missing clinical data. The proposed system enables early, personalized breast cancer risk assessment and supports preventive decision-making. Its flexibility allows deployment as a Clinical Decision Support System (CDSS) for both healthy individuals and those under diagnostic investigation, contributing to enhanced screening strategies and reduced unnecessary procedures. By addressing the gaps between long-term risk prediction and near-term diagnostic accuracy, the model lays the foundation for a more inclusive and precision-oriented prevention framework.