Model Machine Learning yang Dioptimalkan untuk Prediksi Penyakit Jantung Menggunakan R Shiny

Abstract

Heart disease continues to be a major contributor to global mortality, highlighting the critical importance of early detection in enhancing patient outcomes. The increasing availability of structured clinical datasets has enabled the application of intelligent systems for risk prediction and diagnostic support. In this paper, the effectiveness of three supervised learning algo- rithms—Random Forest (RF), Support Vector Machine (SVM), and Decision Tree (DT)—is evaluated for the task of heart disease prediction. This investigation is based on the Heart Failure Prediction dataset sourced from the Kaggle platform. The training process for each model involved a 10-fold cross- validation, with its hyperparameters later being tuned using grid search optimization. Model efficacy was measured against standard classification benchmarks, including accuracy, sensitivity, specificity, and the area under the ROC curve (AUC). The Random Forest model emerged as the most effective, demon- strating superior performance with an AUC of 0.9517, sensitivity of 81.18%, and specificity of 90.44%. To facilitate clinical use, this model was subsequently integrated into a user-friendly web tool built with the R Shiny framework. The interface allows users to input patient-level clinical data and obtain real-time predictions, along with visualizations of feature importance and risk probability. This implementation bridges the gap between algorithm development and practical application, offering a user- friendly decision support tool for early heart disease screening. The findings affirm that machine learning models, when properly tuned and validated, can serve as effective and interpretable tools in clinical decision-making. This work contributes to the advancement of e-health and the integration of AI-driven models into medical workflows