EUVIMEDEUVIMED entdecken
Crossref

Data-driven heart disease prediction through feature selection and hybrid learning

Auxilia Anitha Mary R, T. Ramaprabha

Aposta: Revista de Ciencias Sociales · 19.05.2026 · 24 (112) · e844

DOI: 10.23900/ra.v24i112.844 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Aposta: Revista de Ciencias Sociales, online erschienen am 19.05.2026. Verfasst von Auxilia Anitha Mary R und 1 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Heart disease requires timely and accurate risk prediction to enhance patient outcomes, facilitate early interventions, and support individualized treatment approaches. However, conventional predictive models are usually constrained by heterogeneous data sources, high-dimensional features, redundancy, and instability of prediction performance. To overcome these limitations, the current study proposes a novel hybrid framework that combines multi-source clinical, biochemical, and lifestyle information for accurate and interpretable heart disease risk prediction. CARDI-FUSE (Cardiovascular Data Fusion and Selection Engine) first extracts and selects the most informative features based on the Adaptive Heart Optimization Algorithm (AHOA) to reduce the dimensionality and maintain clinical interpretability. These optimized features are then fed into HEART-NET (Hierarchical Ensemble Attention-based Risk Tracker Network), a two-stream neural network that synchronizes biochemical, clinical, and lifestyle inputs that are heterogeneous in nature using an attention-based fusion mechanism. VITA-VOTE (Variable-Intelligence Triple Adaptive Voting Classifier) is the next enriched feature representation that can be assessed as a hybrid ensemble of XGBoost, Artificial Neural Network (ANN), and K-Nearest Neighbor (KNN) through the application of dynamic weighted voting to achieve high-level robustness and patient-specific predictive power. Experimental evaluation on the UCI Heart Disease and Framingham datasets yields 97.5% accuracy, 97.2% precision, 97.8% recall, 97.5% F1-score, and an AUC of 0.982, outperforming individual base classifiers by a margin of 5–8%. AHOA reduced feature dimensionality from 76 to 21 clinically significant variables, cutting training time by 35% and memory consumption by 30%. These results establish CARDI-FUSE as an interpretable and computationally efficient framework for cardiovascular risk stratification in heterogeneous clinical data environments.

Zitieren

APA

R, A. A. M., & Ramaprabha, T. (2026). Data-driven heart disease prediction through feature selection and hybrid learning. Aposta: Revista de Ciencias Sociales, 24(112), e844. https://doi.org/10.23900/ra.v24i112.844

Vancouver

R AAM, Ramaprabha T. Data-driven heart disease prediction through feature selection and hybrid learning. Aposta: Revista de Ciencias Sociales. 2026;24(112):e844. doi:10.23900/ra.v24i112.844

BibTeX
@article{R2026,
  author = {R, Auxilia Anitha Mary and Ramaprabha, T.},
  title = {Data-driven heart disease prediction through feature selection and hybrid learning},
  journal = {Aposta: Revista de Ciencias Sociales},
  year = {2026},
  volume = {24},
  number = {112},
  pages = {e844},
  doi = {10.23900/ra.v24i112.844}
}
Kostenlos starten →