Novel hybridized learning-based model for heart disease prediction using principal component analysis and deep learning

Abstract

Heart disease (HD) remains a critical health issue globally, leading to high rates of death and requiring complex, expensive diagnostic processes. The extensive influence of heart failure on morbidity and mortality rates highlights the need for precise and timely diagnosis to facilitate effective prevention, early detection, and treatment, ultimately reducing health risks. However, predicting HD accurately and early is a difficult job due to the complexity of medical data, which healthcare professionals must rapidly decipher for effective treatment. Moreover, the variety of criteria used by different feature selection (FS) algorithms complicates the identification of the most effective method. Hybrid approaches that combine multiple FS methods can address the shortcomings of single-method strategies. To solve the challenges of high-dimensionality and performance constraints, a novel hybrid learning-based model (NHLBM) for predicting HD is introduced. Our method uses two specific strategies: principal component analysis and the Deep Autoencoder (DAE). We evaluated the effectiveness of the NHLBM using the public Z-Alizadeh Sani dataset and nine machine-learning techniques. The NHLBM demonstrates remarkable performance improvements, identifying an optimal dimension for HD prediction, refining its representation through DAE, and achieving up to a 9% increase in predictive accuracy. We found that the NHLBM gets an amazing 87.91% accuracy rate when it uses logistic regression with scaling method, optimal parameterization, and a data split ratio of 70:30. Furthermore, the second strategy significantly improved the NHLBM’s accuracy over the first, marking a notable progression in the domain of medical prediction and diagnosis of HD.