Sparse attention FT-transformer for early breast cancer prediction using structured clinical data

Abstract

Early detection of breast cancer is essential for improving clinical outcomes and reducing mortality. This study proposes a Sparse Attention–based FT-Transformer framework for early prediction using structured clinical diagnostic data, designed to capture high-order feature interactions while minimizing redundancy through sparsity-driven attention. The model is evaluated on a publicly available breast cancer dataset derived from fine needle aspiration biopsies, using standard preprocessing and stratified validation techniques. Experimental results demonstrate that the proposed model outperforms conventional methods such as Support Vector Machine, Random Forest, and Logistic Regression, achieving over 99% accuracy along with high precision, recall, and F1-score. These findings highlight the model’s effectiveness as a reliable and interpretable decision-support tool, offering significant contributions to early diagnosis and potential improvements in clinical practice.