A Deep Learning Approach for Cardiovascular Risk Prediction

Abstract

Cardio Vascular Diseases (CVDs) are known to cause high percentage of mortalities in the global population; hence, the necessity of reliable and timely systems of risk prediction has been highlighted in the literature. This paper introduces a new deep learning model, which is called Cardio-RiskNet, to predict cardiovascular risk and offers a rich set of clinical and lifestyle variables as inputs. The model developed in Python was tested on the popular dataset on Hearts Disease Health Indicators that could be obtained on Kaggle. The data includes various information including age, body mass index, blood pressure, cholesterol, glucose level, physical activity, and smoking behaviors. The class imbalance issue is addressed with the SMOTE-ENN hybrid resampling strategy, on the one hand, it gives an opportunity to better identify the minority classes, and on the other, it enhances the model generalization. Cardio-RiskNet is based on the integration of a Conv1D framework with Residual blocks and Squeeze-and-Excitation (SE) modules to perform effective hierarchical feature extraction and powerful representation learning. This was trained over 50 epochs and the model achieved a 84.21 percent accuracy and an AUC of 0.7074 which is very good predictive performance in the field of cardiovascular risk assessment. In addition, the Explainable AI (XAI) methods have been applied to present a more vivid understanding of the decisions made by the model, therefore, enabling the use of the model in a safe and reliable setting in terms of clinical decision support. Altogether, Cardio-RiskNet is an effective, easy-to-understand, and scalable application that will aid in the early prevention of cardiovascular risk, and it is highly likely to be adopted in preventive care and analytics in healthcare.