XGBoost Algorithm Identifies Key Breast Cancer Risk Factors With High Accuracy

A comparative machine learning study evaluating predictive models for breast cancer found that the Extreme Gradient Boosting algorithm achieved the highest classification accuracy among five evaluated models. According to research published in BMC Medical Research Methodology, Extreme Gradient Boosting reached a 97.3 percent accuracy rate on benchmark breast tumor data, while delivering superior numerical results across precision, recall, F1-score, and training time.

The study was designed to predict breast cancer and identify associated risk factors by directly comparing the performance of Extreme Gradient Boosting with four conventional machine learning architectures. To address the interpretability of model outputs, the researchers integrated SHapley Additive exPlanations analysis to explain predictions and clarify the underlying relationships between key predictive variables.

The research team retrieved breast tumor datasets from the University of California, Irvine Machine Learning Repository to construct and evaluate the classification frameworks. Feature selection was conducted using heatmaps of the correlation coefficient matrix, allowing the authors to systematically eliminate redundant and highly correlated variables prior to model training.

Five machine learning models were systematically analyzed: Extreme Gradient Boosting, Random Forest, K-Nearest Neighbors, Decision Tree, and Support Vector Machine. The evaluation protocol employed five-fold cross-validation to assess classification accuracy robustness, generated receiver operating characteristic curves with DeLong tests to compare area under the curve metrics, and applied McNemar’s test for pairwise accuracy comparisons across models.

Confusion matrix results showed that Extreme Gradient Boosting achieved the highest accuracy at 97.3 percent, followed by Random Forest at 95.6 percent, K-Nearest Neighbors at 94.7 percent, Support Vector Machine at 92.1 percent, and Decision Tree at 89.5 percent. Receiver operating characteristic analysis showed area under the curve values of 0.97 for Extreme Gradient Boosting, 0.96 for Random Forest, 0.94 for K-Nearest Neighbors, 0.93 for Decision Tree, and 0.90 for Support Vector Machine, although the authors noted no statistically significant difference in overall performance between Extreme Gradient Boosting and Random Forest.

Interpretability assessments using SHapley Additive exPlanations revealed that radius-worst, concave points-worst, and concavity-worst served as the primary predictive factors linked to high breast cancer risk. In addition, the analysis detected a significant interactive correlation between the radius-worst and concave points-worst features, identifying radius-worst as a pivotal indicator associated with breast cancer occurrence.

Subject of Research: Machine learning prediction modeling and SHAP interpretability analysis for breast cancer diagnosis

Article Title: Breast cancer prediction modeling based on SHAP interpretability analysis and XGBoost algorithm

Article References: Guan, X., Cui, J., Bai, L., Bi, X., Liu, Y., Ren, C., Wang, Z., & Li, S. (2026). Breast cancer prediction modeling based on SHAP interpretability analysis and XGBoost algorithm. BMC Medical Research Methodology. https://doi.org/10.1186/s12874-026-03087-6

Image Credits: AI Generated

DOI: 10.1186/s12874-026-03087-6

Keywords: breast cancer prediction, XGBoost, SHAP analysis, Random Forest, K-Nearest Neighbors, Support Vector Machine, Decision Tree, UCI Machine Learning Repository, radius-worst, concave points-worst, concavity-worst, ROC curve, DeLong test, McNemar's test, five-fold cross-validation, feature selection, classification accuracy, AUC