A TabNet-Based Deep Learning Approach for Cardiovascular Disease Prediction

Abstract

Cardiovascular diseases (CVDs) remain the predominant cause of global mortality, responsible for around 17.9 million deaths per year. Accurate and prompt forecasting of cardiovascular disease risk is crucial for reducing mortality and enhancing patient treatment. This article presents a TabNet-Based deep learning system for predicting cardiovascular disease (CVD) utilizing advanced preprocessing techniques (Winsorization, Box-Cox, Normalization) and clinically pertinent feature engineering (BMI, MAP, PP). The model attained modest predictive efficacy (Accuracy 73%, ROC-AUC 0.73). Feature importance analysis identified cholesterol, systolic blood pressure, and age as the primary predictors. The paradigm, although interpretable, necessitates additional validation across several datasets to establish therapeutic usefulness. The proposed TabNet system efficiently amalgamates preprocessing, feature engineering, and explainable deep learning to enhance the early identification of cardiovascular diseases and promote transparent clinical decision-making. The use of domain-driven features improves predictive reliability and medical interpretability, establishing a robust basis for future integration into clinical decision support systems.

Ähnliche Arbeiten