سامانه همایش‌ها TSTA | ثبت‌نام و ارسال مقاله کنفرانس‌ها و کنگره‌های ملی

Home / سی و دومین کنفرانس ملی و دهمین کنفرانس بین المللی مهندسی زیست پزشکی ایران

Comparative Evaluation of Feature Selection Techniques for Six-Month Mortality Prediction in Heart Failure Patients

Authors :

Parsa Haghighatgoo¹ Somayeh Afrasiabi²

1- دانشگاه شیراز 2- دانشگاه شیراز

Keywords :

heart failure،six-month mortality prediction،machine learning،MIMIC-III،feature selection،XGBoost،SMOTEENN

Abstract :

Heart failure (HF) is a leading cause of hospitalization and mortality worldwide. Accurate early prediction of six-month mortality can support timely and informed clinical decision-making. This study presents a reproducible machine learning pipeline for predicting six-month mortality in HF patients using the MIMIC-III critical care database. We systematically evaluated multiple feature selection methods to identify the most informative features for an XGBoost classifier. Model performance was assessed using recall and ROC-AUC scores within a consistent 10-fold cross-validation framework. Among the tested methods, the L1-based selector achieved the highest performance, with a recall of 0.738 and a ROC-AUC of 0.678. Beyond performance benchmarking, analysis of the selected features revealed both convergence and diversity across methods. Consistently identified predictors included age, BNP, creatinine, BUN, sodium, hemoglobin, and LVEF, all of which are well-established markers of HF prognosis. Comorbidities such as atrial fibrillation, hypertension, and chronic renal failure were frequently highlighted by Boruta and mutual information, while SHAP emphasized renal markers (creatinine, WRF), BNP, and hemoglobin, aligning closely with clinical evidence. These findings demonstrate that the proposed pipeline not only improves model performance but also yields clinically interpretable insights that are in agreement with established HF risk factors.