Real-Time Cardiovascular Risk Prediction Using Interpretable Deep Learning

Abstract

Cardiovascular diseases (CVDs) remain one of the leading causes of global mortality, necessitating accurate and early risk prediction to support timely clinical interventions. Although machine learning and deep learning approaches have shown promise for cardiovascular disease prediction, existing studies often lack adequate temporal feature modelling, insufficient generalisation analysis, and insufficient comparative evaluation across different learning paradigms. To address these limitations, this study proposes a hybrid deep learning framework that combines one-dimensional Convolutional Neural Networks (1D-CNN) and Long Short-Term Memory (LSTM) networks to predict cardiovascular risk from sequential health data. The proposed architecture leverages the 1D-CNN's feature-extraction capability to capture local spatial patterns, whereas the LSTM component models long-term temporal dependencies inherent in physiological signals. The model was evaluated under multiple train–test split configurations (60–40, 70–30, 75–25, and 80–20) to assess its robustness and generalization. The performance was benchmarked against a Dense Neural Network and a Random Forest classifier using comprehensive evaluation metrics, including accuracy, precision, recall, F1 Score, and ROC AUC. The experimental results demonstrate that the LSTM+1D-CNN model achieves consistently high predictive performance, with an accuracy exceeding 93% and F1-scores above 0.95 across most data splits. Comparative analysis shows that the proposed hybrid model offers superior temporal learning and balanced precision recall trade-offs compared with traditional machine learning methods. The training and validation loss curves further indicated stable convergence and minimal overfitting, reinforcing the reliability of the proposed approach. Overall, this study addresses critical research gaps identified in the existing literature by integrating temporal modelling, robust validation, and comparative analysis, thereby contributing a reliable and scalable deep learning framework for cardiovascular disease prediction. These findings highlight the potential of hybrid deep learning architectures for advancing data-driven cardiovascular healthcare systems.