Optimized Machine Learning Models for Heart Disease Prediction: A Performance Analysis

This paper presents a performance analysis of optimized machine learning (ML) models used to predict heart disease. The primary aim of this research was to enhance the accuracy and efficiency of early detection and treatment. Our study employed 6 machine learning (ML) models: Discriminant Analysis, Naïve Bayes, Support Vector Machines (SVM), K-Nearest Neighbors (KNN), Ensemble, and Neural Networks (NN). These models were fine-tuned using hyperparameter tuning techniques. The dataset used in this study was the Heart Disease dataset provided on the Kaggle website. The dataset was split into train-test data configurations of 90-10, 85-15, 80-20, 75-25, and 70-30 in percentage for model learning and performance evaluation.Our study demonstrates that optimizing the hyperparameters of Support Vector Machines (SVM) resulted in highly accurate predictions of heart disease. This model outperformed other machine learning models across various train-test dataset splits, particularly at configurations of 90-10, 85-15, 80-20, and 75-25. Notably, the model achieved the highest accuracy of 95% in the 85-15 train-test dataset split configuration. These findings offer valuable insights for healthcare professionals, equipping them with reliable tools for early heart disease detection and well-informed decisionmaking. It is crucial to emphasize that these machine learning models should complement decision-making and should not substitute professional medical judgment.