<![CDATA[Heart disease is one of the leading causes of death globally, making the development of accurate classification models based on clinical data essential to support early risk stratification. The Decision Tree algorithm is widely applied in medical analysis due to its interpretability; however, its performance is often limited by the use of default hyperparameters. This study aims to improve the performance of the Decision Tree algorithm through the application of hyperparameter optimization using a two-stage strategy. Experiments were conducted using a Kaggle dataset consisting of 918 patients with 12 clinical attributes. The data preparation stage included encoding categorical variables and evaluation using stratified 10-fold cross-validation. The baseline Decision Tree model achieved an accuracy of 79.20%, precision of 83.16%, recall of 78.76%, and an F1-score of 80.68%. The two-stage optimization involved Random Search cross-validation to explore the parameter space, followed by refinement using Grid Search cross-validation. The optimized model showed improved performance, achieving an accuracy of 83.66%, precision of 84.17%, recall of 86.42%, and an F1-score of 85.13%. To test the statistical significance of the performance improvement, a Shapiro-Wilk normality test was conducted on the difference in F1-scores, indicating a normal distribution (p = 0.233). A paired t-test showed that the increase in F1-score was statistically significant (t(9) = 4.60, p = 0.0016) with a very large effect size (Cohen’s d = 1.45).]]>
Copyrights © 2026
