A SURVEY OF PERFORMANCES OF SOME SELECTED MACHINE LEARNING ALGORITHMS FOR CARDIOVASCULAR DISEASE PREDICTION

Abstract

Cardio-Vascular Diseases (CVDs) are the leading causes of early deaths in the world. Middleand
low-income countries suffer the biggest challenge of effective diagnosis and treatment due
to the inadequacy of efficient diagnostic tools and physicians. This affects the proper prediction
and treatment of patients. Though, large proportion of CVDs could be prevented but they
continue to escalate mainly because preventive measures put in place are inadequate. Huge
CVD data is available in the healthcare sector which led to several researches. The University
of California, Irvine (UCI) heart disease data has been used extensively by machine learning
researchers in trying to come up with a more efficient predictive model. Previously, the focus
was on investigating the performances of some selected machine learning algorithms on the
UCI data. These algorithms include Naïve Bayes (NB), Support Vector Machine (SVM),
Decision Tree (DT-J48), Random Forest (RF), K-Nearest Neighbor (KNN) and Artificial
Neural Network (ANN). There are few researchers who used CVD datasets other than that of
the UCI. This paper carries out a survey on the performances of these algorithms on CVD
prediction using 12various datasets other than the UCI data. From our investigation on the 18
researches conducted, most of them in 2018 and 2019, we found that DT-J48, NB and SVM
gained much attention than any other algorithm, where J48 was used 11 times and appeared
the most used algorithm for developing clinical decision support systems. NB and SVM
appeared 10 and 9 times respectively, ANN was employed 8 times, while KNN and LR were
considered 3 times each. RF appeared with the least frequency of 1 only. Finally, it has been
discovered that no single algorithm would be generalized as the best in CVD prediction based
on the data in which it was used.