Skip to content
Open access

Machine Learning-Based Multi-Class Prediction of Breast Cancer Tumor Stage Using Clinical and Pathological Features: A METABRIC Cohort Study

Sep 2026 · International Journal of Innovative Science & Technology · 0 citations · 7 references

Abstract

Breast cancer is the most frequently diagnosed cancer among women worldwide, and the tumor stage recorded at diagnosis strongly determines treatment planning, prognosis and survival. Unlike most prior METABRIC-based studies that predict molecular subtype or survival, this research benchmarks thirteen classifiers with four resampling strategies, hyperparameter optimization, cross-validation and SHAP-based explainability specifically for multi-class tumor-stage prediction from routinely collected clinical variables. Clinical and pathological records of 2,509 METABRIC patients are used, retaining 17 attributes and a four-level stage label. Stage 3 and 4 are merged into a single Advanced class to counter sparsity, and thirteen classifiers are trained on an 80:20 stratified split, compared against SMOTE, ADASYN, Borderline-SMOTE and random under sampling, then tuned via grid search and validated with 5-fold cross-validation. Random Forest with ADASYN gave the strongest pipeline, reaching 83.9% accuracy and macro-F1 of 0.770, versus 0.753 without resampling. The tuned model achieved a macro-AUC of 0.943, while a neural network underperformed the tree ensembles (macro-F1 = 0.75). SHAP identified positive lymph nodes, tumor size and the Nottingham Prognostic Index as the leading predictors, consistent with established clinical staging criteria. A resampled, tuned and explained ensemble model can support, but not replace, clinical tumor staging, pending further validation.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.