Heart disease continues to be one of the primary causes of death around the world, and early identification is crucial for lowering death rates and patient outcomes. While traditional diagnostic techniques are effective, but they often take large amount of time, financial investment, and expert analysis. This thesis investigates the capabilities of machine learning algorithms in predicting heart disease, providing a means for diagnostic support that is efficient, precise, and widely accessible. We have utilized several machine learning models, including Logistic Regression, Support Vector Machines, Naïve Bayes, Random Forest, Gradient Boosting, ANN, XGBoost and KNN, on a dataset consisting of patient health records featuring essential factors like age, cholesterol levels, blood pressure, and various lifestyle elements. The research encompassed data preprocessing, feature selection, and model optimization to improve prediction accuracy. To identify the key features of heart disease, seven performance metrics (Accuracy, Classification error, Prediction Time, Precision, Sensitivity, F-Measure and Specificity) are employed, which provide better insight into the behavior of various feature-selection combinations. By analyzing the seven matrices values of the eight models, we have chosen three models (Logistic Regression, Gradient Boosting and Random Forest) from them and we propose a novel method (LGR Model) by combining these three models for getting higher accuracy. The accuracy of the proposed model is 88%.

Logistic Regression Gradient Boosting Random Forest Prediction Heart Disease