Harnessing a Hybrid Dimensionality Reduction Strategy with Temporal Convolutional Networks for Heart Disease Detection

Abstract

Heart disease (HD) remains one of the major causes of global mortality, with efficient management and prevention mainly relying on early identification and precise diagnosis. However, conventional methods struggle to achieve improved HD detection results. The SDG 3 focuses on preventing unnecessary suffering from avoidable diseases and early demise by concentrating on key targets that improve the health of a country's overall population. Recent advances in artificial intelligence techniques enable the detection of diseases in the initial stage and improve survival rates. This paper focuses on an Integrating Feature Reduction Models with Causal Analysis for Heart Disease Prediction (TFRM-CAHDP) approach, that incorporates the following components: (i) Preprocessing, comprising missing data handling, outlier detection, and data normalization, (ii) A hybrid feature selection strategy, integrating Boruta and minimum Redundancy Maximum Relevance (mRMR), (iii) A Temporal Convolutional Network (TCN) to classify heart disease cases accurately, and (iv) Adam optimization for efficient parameter tuning to ensure faster convergence and improve generalization performance. A broad range of experimental analyses on a benchmark dataset demonstrated the improved performance of the TFRM-CAHDP model compared to recent approaches, achieving a maximum accuracy of 97.40%.