L Babitha¹ , Krishnan K Sundara²

Abstract

Purpose: To apply Hybrid Feature Deep Learning (HFEDL) and structured feature-selection framework (Selected Features) model to develop algorithm for enhanced preventive cardiovascular disease risk prediction. Methods: The HFEDL model is a hybrid feature-engineering and deep-learning method that improves learning dynamics, whereas the structured feature-selection framework is based on statistically significant features to achieve higher classification accuracy. Benchmark datasets have been used to compare various models, including traditional deep learning and the random forest classifier. Results: The HFEDL model achieved 97.5 % testing accuracy and 0.2 loss, the quickest convergence period (15 epochs), and generalization. The structured feature-selection framework model was also able to achieve 90.8 %, and the model had a steady performance. The models generated significant outcomes that were better in terms of accuracy, training efficiency, and generalization (5/5) when compared to traditional state-of-the-art models; thus, the models may be considered to offer real-time clinical decision support. The predicted risk levels support therapeutic decision-making by guiding pharmacological interventions such as statins, antihypertensive therapy, and patient monitoring strategies. Conclusion: The HFEDL model has better predictive capability than traditional machine learning and isolated deep learning models, and the feature-selection framework is more interpretable and less complex to calculate because it identifies clinically meaningful variables. Combining feature selection with hybrid deep learning constitutes a trade-off between accuracy, efficiency, and interpretability.