Stacking ensemble-based breast cancer classification: Enhancing diagnostic accuracy with deep learning and real-time web deployment

Breast cancer remains one of the most prevalent and life-threatening diseases, requiring early and accurate diagnosis to improve survival rates. Traditional diagnostic methods rely on manual interpretation of ultrasound and histopathology images, which are time-consuming, prone to variability, and dependent on expert radiologists and pathologists. Recent advances in deep learning have shown promise in automating breast cancer detection; however, existing models often suffer from overfitting, dataset biases, and poor generalization across different imaging modalities. To address these challenges, we propose a novel stacking ensemble-based breast cancer classification model integrating EfficientNetB8, RegNet, RepVGG, and MNasNet. Our approach enhances classification robustness by leveraging complementary feature extraction capabilities of multiple architectures. We evaluate our model on two publicly available datasets—BUSI (ultrasound) and BreaKHis (histopathology)—demonstrating superior performance over previous deep learning approaches. Our ensemble model achieves a maximum MCC of 99.31% on the BUSI dataset and 99.52% on the BreaKHis dataset, outperforming individual architectures. Additionally, we incorporate Contrast Limited Adaptive Histogram Equalization for contrast enhancement and employ data augmentation to mitigate class imbalance and improve model generalization. Furthermore, we develop a web-based diagnostic system for real-time breast cancer classification, enabling efficient and accessible clinical decision-making. While the proposed approach significantly enhances classification accuracy, future research will focus on dataset expansion, real-world validation, and explainable AI integration for improved interpretability and clinical adoption.