JASIC Volume. 3, Issue 1 (2022)

  1. Home
  2. Volume. 3, Issue 1

Contributor(s)

Olanloye Odunayo, Olawumi Olasunkanmi, Adetoye Adeyemo, Adebayo Segun
 

Keywords

Heart Cardiovascular K-Nearest Neighbors Support Vector Machine Naive Bayes Random Forest
 

Download Full-text (PDF)

... Download File [ 1.16 MB ]
 
Go Back

PERFORMANCE EVALUATION OF MACHINE LEARNING ALGORITHMS IN THE DIAGNOSIS AND CLASSIFICATION OF HEART DISEASES

Abstract: Clinical reports and research have established that heart diseases are a typical example of cardiovascular disease that has sent millions of people globally to an untimely grave. World Health Organization (WHO) also confirmed this assertion and as a result, there have been series of attempts by researchers in various fields to solve this problem. Certain researchers in computer and health informatics carried out predictive analytics to detect and classify the disease based on several biomarkers identified in the affected individual. Meanwhile, enough has not been done to determine the level of susceptibility of individuals to heart diseases with concerted effort on the key indicators such as age, sex, sugar level and some other related attributes before predictive analytics are made. This explores the attribute and it was finally established that sex, age, level of cholesterol etc. are strong markers to determining the level of susceptible of patient to heart disease. Moreover, Four ML models - KNN, NB, SVM and RF were implemented and evaluated in term of their performances in the classification of heart diseases using cross-validation and test dataset. At first, with every feature available in the dataset and later with only the correlated features identified in the descriptive analytics. It was established that accuracy improves across all models when only correlated features were used and SVM exhibits the highest accuracy and F1 Score (84%). Therefore, SVM performs better than KNN, RF and NB when all the models were evaluated on the 25% test set of the correlated features. It could be therefore concluded that in-depth understanding of features for identification of strong disease biomarkers will enhance more accurate diagnostics and this in turn will be of great assistance to the medical practitioners and other stake holders to track susceptibility of individuals with identified features to heart disease
 
 
 

Journal of Applied Science, Information and Computing (JASIC), 2024