ARCHIVES

Heart disease remains one of the most life-taking diseases across the world, and thus enormous research work is geared toward the establishment of strong models for early-stage detection and risk prediction. This paper presents a new hybrid machine learning framework, including deep features obtained via autoencoders, self-adaptive feature recalibration, and ensemble learning to develop an efficient heart disease classification system. It considered a dataset from five established heart disease datasets. The autoencoder captures high-level representations of the data, while the recalibration mechanism is employed to dynamically adjust the importance of the features for the optimization of performance by the proposed model. Multiple classifiers are integrated using ensemble learning-based RF, Extra Trees, and XGBoost (eXtreme Gradient Boosting) aggregators to make the system robust and generalize the results. The proposed hybrid model achieved an accuracy rate of 92.45%, significantly higher than the rates achieved by traditional models and the ensemble techniques of previous studies. Moreover, it yielded very high sensitivity of 93.2% and specificity of 91.4%, hence its effectiveness in identifying high- risk patients while keeping false positives at a minimum. It has much potential for early diagnosis and decision-making in heart disease management. Future research will involve the validation of this model across different populations and optimization for real-time clinical use.