EVALUATION OF DECISION TREES, NAÏVE BAYES, AND SUPPORT VECTOR MACHINES IN FORECASTING CARDIOVASCULAR DISEASE AND FEATURE SELECTION USING THE CHI-SQUARE (Χ²) TEST IN TWO DIFFERENT DATASETS

Abstract

Purpose: Heart disease is a major global concern. In all its manifestations, heart disease is one of the world's leading causes of death. Affordable treatment and early detection are essential for prevention. The correct diagnosis procedure is crucial for every patient. The death ratio is decreased with early detection and prognosis of heart disease.  A number of data mining techniques, including Naive Bayes, decision trees, and support vector machines, helps to detect cardiovascular diseases early and accurately.

Methods: Certain data mining approaches, such as Naive Bayes, decision trees, and support vector machines, aid in the accurate prediction of cardiovascular illnesses using two different datasets. Similar to the ‘Heart Statlog Cleveland Hungary Final’ heart dataset, which has 1190 instances (rows), 11 input characteristics, and 1 output attribute, the ‘Heart Failure Clinical Records’ cardiac dataset has 299 instances (rows), 12 input attributes, and 1 output attribute. Both datasets are taken from UCI machine learning repository.

Results: Using Decision Tree, Naïve Bayes, and SVM classifiers, respectively, the ‘Heart Failure Clinical Records’ heart dataset yielded 92%, 81.67%, and 82.67% prediction accuracy. Similarly, ‘Heart Statlog Cleveland Hungary Final’ heart dataset —achieved 85.23%, 81.88%, and 81.879% prediction accuracy. Finally, the longest paths of all attributes in these two datasets are determined using decision trees.

Conclusion: After determining the effectiveness of three data mining algorithms—Decision Tree, Naïve Bayes, and Support Vector Machine—for two heart datasets, the chi-square test is used to select features in descending order that are strongly associated with the presence of heart disease.