Machine Learning–Based Diagnostic Classification of Active Breast Cancer Using Clinical and Blood-Based Biochemical Biomarkers

Background: Breast cancer remains the most frequently diagnosed malignancy among women globally, with rising incidence in low- and middle-income countries. Conventional screening methods, such as mammography, ultrasound, and magnetic resonance imaging (MRI), have limitations related to cost, accessibility, and diagnostic accuracy, highlighting the need for complementary strategies. This study aimed to improve the diagnostic classification of prevalent, undiagnosed breast cancer by integrating routine clinical variables with blood-based biochemical markers using machine learning models. Methods: A retrospective cohort of 2000 women was analyzed from January to December 2024. Clinical parameters (age, menopausal status, family history, comorbidities) and routinely measured biomarkers (hemoglobin, red cell distribution width, platelet distribution width, mean corpuscular hemoglobin concentration, C-reactive protein, lipid profile, and fasting glucose) were included. Logistic regression, decision trees, random forests, extreme gradient boosting (XGBoost), and neural networks were trained and evaluated using stratified cross-validation. Model performance was evaluated using accuracy, sensitivity, specificity, F1-score, the receiver operating characteristic (ROC) curve, and the area under the curve (AUC). Feature importance was examined using Shapley additive explanations (SHAP) values to enhance interpretability. Results: XGBoost achieved the highest diagnostic performance (accuracy 81.7%, AUC 0.86), followed closely by neural networks (accuracy 80.5%, AUC 0.85). The random forest also performed strongly (accuracy 78.0%, AUC 0.82). Logistic regression and decision tree models yielded lower accuracies of 71.0% and 69.2%, respectively. SHAP analysis identified platelet distribution width, age, red cell distribution width, hemoglobin, and C-reactive protein as the most influential predictors. The integrated model achieved an AUC of 0.86, compared with 0.77 for the clinical-only model and 0.78 for the biomarker-only model. Conclusion: Machine learning models integrating clinical and biochemical data provide accurate, noninvasive, and scalable diagnostic classification of active breast cancer. By complementing imaging-based screening, such tools may support early identification of active breast cancer, particularly in resource-limited settings. Further validation in diverse populations and prospective clinical studies is warranted.