Enhancing Breast Cancer Detection: FT-Transformer vs. Traditional Models

Breast cancer remains one of the most prevalent and life-threatening diseases worldwide, necessitating accurate and early diagnosis to improve patient outcomes. In this study, we explored the application of machine learning (ML) and deep learning (DL) models for breast cancer classification using the Breast Cancer Wisconsin Diagnostic Dataset. We implemented and compared three models: Random Forest (RF), XGBoost, and FT-Transformer, evaluating their performance based on accuracy, recall, precision, and AUC-ROC score. The experimental results demonstrated that FT-Transformer significantly outperformed the other models, achieving an accuracy of 93.2%, recall of 93.7%, and an AUC-ROC score of 95.1%, making it the most reliable model for tumor classification. The model’s self-attention mechanism enabled it to effectively capture complex feature dependencies, leading to improved classification confidence and reduced false negatives. SHAP feature importance analysis revealed that tumor concavity, texture, and smoothness were the most influential features in classification. Additionally, the AUC-ROC analysis confirmed FT-Transformer’s superior discriminatory ability, ensuring a high true positive rate with minimal false positives. This study highlights the potential of deep learning-based models in medical diagnosis, emphasizing their role in enhancing early cancer detection and clinical decision-making.