Early breast cancer prediction using optimized machine learning and tumor-immune modeling

Abstract

This study aims to enhance early breast cancer prediction accuracy by utilizing machine learning classifiers and feature selection techniques. The Wisconsin Diagnostic Breast Cancer (WDBC) dataset was used to train and evaluate three popular machine learning classifiers: Support Vector Machine (SVM), Random Forest (RF), and k-Nearest Neighbors (k-NN). Feature selection methods were applied to optimize model performance, including Recursive Feature Elimination (RFE) and Mutual Information. Cross-validation and hyperparameter tuning were conducted to ensure the robustness and reliability of the models. The results showed that the SVM classifier achieved the highest performance with an accuracy of 98 %, compared to 95.8 % for RF and 96.2 % for k-NN. The SVM model demonstrated a precision of 0.98 and a recall of 0.95 for malignant cases. Feature selection revealed that mean radius, texture, and area were the most influential features, and SHapley Additive exPlanations (SHAP) analysis confirmed their clinical relevance in breast cancer diagnosis. A tumor-immune dynamic model also indicated that treatment efficacy (γ = 0.0500/ day) was a critical parameter for tumor control. Statistical significance tests (p < 0.05) confirmed that the SVM classifier outperformed the other models. This study highlights the potential of combining machine learning with clinical insights to develop an effective framework for breast cancer prediction, offering high diagnostic accuracy and biological interpretability.