A Hybrid Framework for Breast Cancer Detection Using Rule Extraction, Deep Learning and Ensemble Learning Methods for Improved Interpretability
Abstract
Breast cancer remains a pressing health crisis in Nigeria. It becomes crucial to consider interpretable AI solutions for early detection. This study develops a hybrid framework for detecting breast cancer prevalence in patients. The methodology of this study encompasses normalization, augmentation, 512D convolutional feature extraction on 7,632 breast cancer mammogram images, combining CNN+ RF+G-REX for transparent interpretable hybrid model capable of generating decision rules, addressing the black-box opacity of deep learning and ensemble learning models highlighted in prior works. The model was evaluated on accuracy, precision, recall, F1-score, and AUC-ROC. The hybrid achieves superior balanced performance (90.22% accuracy, 93.63% precision, 86.30% recall, 89.82% F1, 97.22% AUC-ROC), outperforming standalone CNN (84.74%) and rivalling RF (93.74%) while enabling clinical trust through human-readable explanations. These contributions advance explainable model in medical imaging, offering a lightweight, deployable system for low-resource settings to reduce mortality through enhanced interpretability and diagnostic confidence.