Prediction of Heart Disease using Hybrid Feature Selection

Abstract

Heart disease is serious disorder which threatens many people’s lives and illness in the world’s population. Predicting heart disease helps physicians and doctors to make effective decisions with respect to the health of the patients. Hence the development of machine learning (ML) leads the major part in predicting presence or absence of various serious health disorders. This study seeks to predict heart disease by using various ML models that employ hybrid feature selection. The hybrid selection involves selecting the predictive featuresby applying the fusion of filter-based feature selection and wrapper-based feature. Grid Search approach is then utilised to tune hyperparameters of classification algorithms. Finally, the comprehensive investigation of five ML classifiers such as Decision tree (DT), Logistic Regression (LR), SVM, Random Forest (RF) and Ada boost algorithms are accompanied by using metrics such as confusion matrix, accuracy score, precision, recall, kappa score, F1-score and Receiver operating curve (ROC). In the Kaggle heart disease dataset, this study discovered that the RF technique obtains an accuracy of 91% and recall of 95% compared with other classifiers with reduced feature set.