Predictive Modeling of Heart Failure Using Machine Learning Algorithms: An Empirical Study

Abstract

Heart failure is still one of the world's top causes of death; therefore, it's important to identify high-risk patients early and accurately. To predict heart failure, this research compares nine machine learning (ML) methods using a clinical dataset. The study comprises preprocessing, correlation analysis, mutual information-based feature selection, model training, and performance evaluation utilizing thresholdindependent metrics (Area Under the Receiver Operating Characteristic Curve; AUC, Average Precision Score; APS, Cohen's Kappa, Matthews Correlation Coefficient; MCC) as well as threshold-based metrics (accuracy, precision, recall, F1score). With the highest <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{F 1}$</tex>-score, competitive recall, and precision, the results show that overall K-Nearest Neighbors (KNN) performed the best. The highest precision results were obtained using Logistic Regression (LR); however, ensemble models such as Random Forest (RF) and Gradient Boosting (GB) also showed strong performance. For a thorough diagnostic evaluation, visualization methods such as threshold sensitivity curves, confusion matrices, and mutual information plots were used. This study highlights how interpretable ML models can be used to forecast the risk of heart failure in clinical decision support.

Ähnliche Arbeiten