Attention-Based Explainable AI Framework for Real-Time Cardiovascular Risk Prediction Using Wearable Sensor and Multimodal Healthcare Data

Authors

  • Er. Mandeep Kaur
  • Dr. Khushboo Bansal
  • Dr. Sharanpreet Kaur

Keywords:

Cardiovascular, Heart disease, Artificial intelligence, Explainable deep learning

Abstract

Among other illnesses, cardiovascular diseases have been listed among the top causes of death globally. Therefore, an intelligent and explainable healthcare system can play an integral role in predicting cardiovascular diseases and continuously analyzing patients' conditions. In this context, this research work provides a solution for the classification of patients into Low Risk, Moderate Risk, and High Risk of developing cardiovascular diseases based on the real-time cardiovascular disease prediction model incorporating BiLSTM, attention mechanism, and SHAP-based XAI. In this regard, the proposed model is designed to use multi-modal health care data such as electrocardiography (ECG)-related health care parameters, blood pressure, cholesterol, heart rate, and angina triggered by physical activity. Preprocessing of healthcare data was initially done through removing missing values from the dataset and normalization of the healthcare features using the min-max technique. Extraction of sequential healthcare patterns and assigning more relevance to clinically important cardiovascular features is done by BiLSTM networks and attention mechanism, respectively. The preliminary experiment performed on a clinical cardiovascular dataset, which consisted of 303 patient samples and 14 healthcare features, indicated that the designed BiLSTM-Attention architecture obtained 100% and 41.67% accuracy scores during training and testing phases, respectively. In order to increase the accuracy and avoid overfitting, an increased cardiovascular dataset composed of around 70,000 patient samples are employed for further experimentation. As a result, the modified algorithm showed great success in improving predictive ability and produced a score of about 89.77% accuracy. The SHAP analysis suggested that chest pain type, cholesterol, heart rate, and angina induced by physical exercise were key cardiovascular risk factors affecting the output of the prediction model.

Downloads

Published

2026-11-16

How to Cite

Kaur, E. M., Bansal, D. K., & Kaur, D. S. (2026). Attention-Based Explainable AI Framework for Real-Time Cardiovascular Risk Prediction Using Wearable Sensor and Multimodal Healthcare Data. Adolescência E Saúde, 1076–1083. Retrieved from https://adolescenciaesaude.com/index.php/aes/article/view/3460

Issue

Section

Original Articles