Enhancing Breast Cancer Diagnosis: Stacked Neural Networks for Reducing Data Imbalance

Among female malignancies, breast cancer is still rather frequent globally. Mammography, an essential diagnostic tool, provides a non-invasive means of early detection. Machine learning models can aid in automating breast cancer detection, but they rely heavily on high-quality, balanced datasets. This research introduces an architectural framework, 'Stacked Neural Network,’ addressing data imbalance in multiclass medical datasets, particularly for breast cancer diagnosis. Emphasizing the impact of imbalanced data on model performance in critical domains like medical diagnosis, the paper advocates for equitable predictions through data balance. Utilizing the Wisconsin Diagnostic Breast Cancer dataset, a Stacked Neural Network model is trained and evaluated. Data pre-processing involves cleaning, encoding, and partitioning for training and testing. The architecture comprises ReLU activation, dropout for overfitting mitigation, and a sigmoid output for binary classification. Results demonstrate impressive accuracies: 98.51% in training and 97.2% in validation for breast cancer classification. The study highlights the model's robustness in distinguishing tumor types