Hybrid Deep Learning with Explainable AI for Breast Cancer Prediction Using the WBCD Dataset

Abstract

Breast cancer remains a leading cause of morbidity and mortality among women worldwide, creating demand for diagnostic models that are both accurate and interpretable. While machine learning and deep learning methods have achieved strong results on benchmark datasets, such as the Wisconsin Breast Cancer Diagnostic (WBCD) dataset, limitations including class imbalance, overfitting, and lack of interpretability hinder their clinical use. This study proposes a hybrid deep learning framework that integrates Convolutional Neural Networks (CNN) with Gated Recurrent Units (GRU), supported by class weighting and stratified sampling to address class imbalance, and SHAP (SHapley Additive exPlanations) to enhance model transparency. Preprocessing involved KNN imputation for missing values, label encoding, and Min-Max normalization. The model was evaluated using accuracy, precision, recall, F1-score, and AUC-ROC. On the WBCD dataset, the CNN–GRU achieved 95.61% accuracy, 100.0% precision, 88.10% recall, 93.67% F1-score, and 0.9944 AUC-ROC. The absence of false positives underscores the model’s conservative boundary, though five false negatives highlight sensitivity trade-offs. SHAP-based global and local explanations identified concave points, perimeter, and radius as the most influential features, consistent with clinical knowledge and providing insight into both correct and incorrect classifications. Unlike prior studies that reported higher accuracies but offered limited interpretability, the proposed framework strikes a balance between predictive strength and clinical trustworthiness. These findings position CNN–GRU with explainable AI as a promising tool for decision support in breast cancer diagnosis and point to future directions in multimodal integration, federated learning, and prospective clinical validation.