HypGB: High Accuracy GB Classifier for Predicting Heart Disease With HyperOpt HPO Framework and LASSO FS Method

Abstract

Early prediction of cardiovascular disease is crucial for medical experts to make informed decisions. Effective diagnosis of heart disease can help prevent heart failure, heart attacks, stroke, and coronary artery disease. This paper aims to build a high-accuracy heart disease prediction system using machine learning. For this purpose, an automatic machine learning system called HypGB was developed. HypGB uses a Gradient Boosting (GB) model to classify patients with heart disease. It also uses a standard LASSO feature selection technique to identify the most informative feature subset and remove noisy and redundant features from clinical data. The GB model was also optimized with the latest HyperOpt optimization framework to determine the best configuration of the hyperparameters. Experimental results using two open-source heart disease clinical datasets (Cleveland heart disease and Kaggle heart failure) indicate that HypGB identifies the most accurate features and obtains the optimal combinations of hyperparameters that efficiently predict heart disease. It achieved the highest classification accuracies of 97.32% and 97.72% using the Cleveland and Kaggle datasets, respectively, which outperformed the previous approaches. With the highest accuracy, the HypGB system shows its potential for implementation in the healthcare domain to help medical experts predict heart disease fast and accurately.

Keywords

• Cardiovascular disease
• machine learning algorithms
• hyperparameter optimization
• redundant features
• disease prediction
• clinical data analysis