A Decision Tree – Based Heart Disease Prediction Model with SMOTE and Fractional-Order Feature Transformation

This study compares two machine-learning pipelines for heart disease prediction using lipid-profile variables: Decision Tree with Synthetic Minority Oversampling Technique (SMOTE) and fractional-order feature transformation followed by Decision Tree with SMOTE. A truncated Grünwald–Letnikov-style transformation with fractional order α = 0.5 was used to construct an alternative feature representation before classification. The study evaluates accuracy, balanced accuracy, precision, sensitivity, F1-score, ROC-AUC, training time, and Decision Tree complexity. The conventional Decision Tree + SMOTE model achieved higher predictive performance on the analyzed dataset, with an accuracy of 87.96% and ROC-AUC of 0.922, whereas the fractional-order model achieved lower predictive values but a shallower tree and slightly shorter training time. The findings indicate that the fractional-order transformation changed the feature representation and tree structure but did not provide a predictive advantage under the experimental conditions considered. The contribution is therefore an empirical comparison of fractional-order feature representation with a conventional SMOTE-based Decision Tree pipeline for lipid-profile-based heart disease prediction.