Answers are grounded in this paper and cite their sources.
A data-modality-driven cross-model classification using machine learning, deep learning, and transfer learning with transparency analysis for breast cancer binary classification across two different datasets is explored in this work. The ML track evaluates five models: Logistic Regression, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Decision Tree. The Wisconsin Breast Cancer Diagnostic dataset, comprising 569 samples and 30 numerical features, is used for this evaluation. The DL track trains three convolutional architectures a baseline CNN, EfficientNetB0 with transfer learning, and DenseNet121 with two-phase fine-tuning on the IDC Regular Breast Histopathology Image dataset. The preprocessing pipeline for ML includes log transformation, stratified train/test splitting, and Standard-Scaler normalization to prevent data leakage. In addition, the DL pipeline incorporates balanced class sampling, a three-way 80/10/10 split; data augmentation; class weighting; and model-internal pixel rescaling. Empirical results demonstrate the effectiveness of the proposed framework, achieving 98.24% accuracy on the WDBC dataset and 91.2% accuracy on the IDC histopathology image dataset. The findings indicate strong predictive capability across both tabular and image-based breast cancer classification tasks. Furthermore, Explainable Artificial Intelligence (XAI) techniques, including SHAP, LIME, and Grad-CAM, were integrated to enhance model transparency and interpretability, enabling better understanding of feature importance and decision-making processes in breast cancer prediction. The study highlights that the adaptability of the paradigm should be guided by data modality rather than blanket assumptions about model superiority.
No takes yet. Share an insight, caveat, or question.
Tiwari et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: