Cardiovascular disease prediction using machine learning algorithms with Bernstein-type kernel functions
摘要
Cardiac disease is still one of the world’s top causes of mortality. After coronavirus disease 2019, for those infected with the coronavirus, the risk of cardiac diseases such as arrhythmias, heart failure and chronic cardiovascular illnesses has increased dramatically. Since, early detection and prevention of the disease depend on precise and effective prediction models. In this field, machine learning (ML) has become a rapidly expanding technique that may be used to find intricate patterns in medical data. This study focuses on heart disease detection using two ML algorithms: support vector machines (SVM) and kernel ridge regression (KR), enhanced through the development of two novel kernel functions obtained from the Bernstein operator and modified Bernstein operator. These kernel functions, constructed to improve the performance of ML algorithms, were integrated into both algorithms to replace traditional kernel functions such as polynomial, linear, sigmoid and radial basis function (RBF). These kernels are extended to multidimensional feature spaces via summation over individual feature-wise Bernstein kernels. The proposed kernels required computationally intensely matrix operations and multifold cross-validation procedures that can best leverage parallel and high-performance computing (HPC) resources for their efficient executions and large-scale deployment. Consequently, using a nested fivefold cross-validation on the Cleveland heart disease dataset, the proposed kernel functions outperformed traditional kernels and some baseline models by achieving the highest mean accuracy of