A Hybrid Transfer Learning Approach Using Obesity Data for Predicting Cardiovascular Diseases Incorporating Lifestyle Factors

Abstract

Cardiovascular Diseases (CVDs), particularly heart diseases, are becoming a significant global public health concern. This study enhances CVD detection through a novel approach that integrates obesity prediction using machine learning (ML) models. Specifically, a model trained on an obesity dataset was used to add an 'Obesity level' feature to the heart disease dataset, leveraging the relation of high obesity with increased heart disease risk. We have also calculated BMI and added as a feature in CVD dataset. We evaluated this transfer learning-based novel approach alongside eight ML models. Performance of these models was assessed using precision, recall, accuracy and F1-score metrics. Our research aims to provide healthcare practitioners with reliable tools for early disease diagnosis. Results indicate that ensemble learning methods, which combine the strengths of multiple models, significantly improve accuracy compared to other classifiers. We are able to achieve a 74% accuracy score along with 0.72 F1 score, 0.77 precision and 0.80 AUC with XGBoost classifier, followed closely by the DNN with 73.7% accuracy with 0.72 F1 score, 0.75 precision and AUC of 0.798 with our proposed model. We seek to enhance healthcare efficiency and promote public health by integrating AI-based solutions into medical practice. The findings demonstrate the potential of ML techniques and the effectiveness of incorporating obesity-related features for optimized cardiovascular disease detection.