Document Type : Research Paper

Authors

• Amirhossein Fallah ¹
• Hannah Pirkhedri ²
• MohammadMehdi Momen ³

¹ Department of Computer Engineering, K. N. Toosi University of Technology, Tehran, Iran.

² Department of Computer Engineering, Bu-Ali Sina University, Hamedan, Iran

³ Department of Computer Engineering, Isfahan University of Technology, Isfahan, Iran

Abstract

Heart failure mortality-outcome prediction requires models that are both clinically sensitive and methodologically re
liable, especially when class imbalance and data leakage can distort performance estimates. This study proposes an
efficient leakage-controlled GGO-XGBoost framework for binary mortality-outcome prediction using the UCI heart
failure clinical records dataset. The outcome is defined by DEATH_EVENT, where class 0 indicates survival during
follow-up and class 1 indicates death during follow-up. The proposed framework combines pipeline-based preprocess
ing, training-only imbalance handling, and stability-aware Greylag Goose Optimization for tuning XGBoost hyperpa
rameters. Logistic Regression, SVM, Random Forest, and baseline XGBoost are used as comparative models under the
same evaluation protocol. On the independent test set, the optimized GGO-XGBoost model achieved AUC 0.8768, re
call 0.7895, F1-score 0.7317, accuracy 0.8167, and precision 0.6818. These results suggest that constrained metaheuristic
optimization can improve clinically relevant classification behavior in imbalanced heart failure outcome prediction, pro
vided that leakage control and conservative model selection are carefully maintained.

Keywords

• Heart failure prediction
• Mortality prediction
• Greylag Goose Optimization
• XGBoost
• Clinical machine learning