Interpretable Heart Disease Detection Model for IoT-Enabled WBAN Systems

Abstract

Heart disease is a leading global health concern, contributing to significant mortality rates. It encompasses a range of conditions affecting the heart, leading to complications such as blocked blood vessels, myocardial infarction, chest pain, and stroke. This study presents an interpretable heart disease detection model specifically designed for Internet of Things (IoT)-enabled wireless body area networks (WBANs). Our approach employs a highway bidirectional gated recurrent unit (BiGRU) network to accurately detect heart disease patients. To enhance the model performance, we address critical data preprocessing challenges, such as outliers in data, class imbalance, and feature selection. We employ a robust scaler data transformation method to mitigate the impact of outliers. The synthetic minority oversampling technique (SMOTE) is applied to address the imbalance in the dataset. We utilize the SelectKBest algorithm with the ANOVA F-test scoring function to select the most relevant features to improve the model efficiency. The dataset is partitioned into training, validation, and testing sets to ensure model generalization. Hyperparameter optimization is performed using a random search method to determine the optimal model architecture. Furthermore, a highway network mechanism is incorporated to enhance information flow, leading to improved training efficiency and detection accuracy. To ensure clinical relevance and acceptability, we employ the SHapley Additive exPlanations (SHAP) technique to provide insights into the model’s decision-making process. Evaluation of unseen test data demonstrates that our proposed model outperforms existing approaches by 1%–9% in terms of detection accuracy.