Breast cancer remains one of the leading causes of death in women across the world, especially considering the vital role of early and accurate diagnosis in disease management. In this paper, a proposal for a hybrid model based on deep learning that combines Convolutional Neural Networks (CNN) with Extreme Gradient Boosting (XGBoost) towards breast cancer detection in mammography images is proposed. Taking advantage of the intrinsic power of CNNs in automatic feature learning and classification capability of XGBoost, the new model is designed to perform better than traditional approaches in diagnostic efficiency. The Augmented INbreast database with thorough preprocessing via normalization and on-the-fly data augmentation is utilized by the paper for inducing dataset diversity as well as overfitting avoidance. A pre-trained CNN model is utilized as a feature extractor and features are passed through an XGBoost classifier for prediction. The experimental result yields much better classification performance with 96.24% test accuracy and satisfactory recall rate in cases of malignancy. The model is good enough for other metrics such as precision, F1-score, and AUC to qualify as a decision-support system for radiologists. This work illustrates the utility of deep learning and ensemble methods for accurate, interpretable, and scalable breast cancer classification from mammography images.