Paramedical Sciences and Military Health

Introduction: Continuous monitoring of heart failure patients using IoT-enabled wearable sensors can improve early diagnosis and reduce mortality. However, imbalanced clinical physiological datasets and suboptimal model tuning limit real-world performance. This
study proposes an IoT-based intelligent health-monitoring framework specifically designed for heart-failure prediction and risk assessment. The proposed system integrates an adaptive deep convolutional neural network (ADCNN) whose weights are optimized by a modified Grey Wolf Optimizer (LF-GWO), together with SMOTE for class balancing, to enhance the accuracy and sensitivity of heart-failure outcome prediction.
Materials and Methods: The framework consists of four phases: data acquisition (wearable/IoT sensors), cloud-edge preprocessing, feature extraction and classification using ADCNN+LF-GWO, and alert generation. Experiments were conducted on the UCI Heart
Failure Clinical Records dataset, containing 299 patient samples with 13 physiological and clinical attributes. Model configuration and preprocessing (including SMOTE) are detailed, and performance was evaluated using standard metrics (accuracy, precision, recall, and confusion matrix) on held-out test sets.
Results: The proposed method achieved an overall test accuracy of 0.9208 (92.08%) and demonstrated high sensitivity in detecting adverse outcomes (up to ≈97% in specific trials). Confusion-matrix analysis on the test split (N=75) yielded 69 correct predictions. A peak accuracy of 99.6% was observed in one iteration, which may indicate model overfitting or instability in that specific run.
Conclusion: Integrating ADCNN with LF-GWO and SMOTE delivers strong predictive performance for heart-failure monitoring and improves sensitivity compared with baseline ML models (e.g., RF, SVM, CNN).
Limitations & Practical implications: External validation on independent datasets was not performed; dataset size is limited, and some reported metrics vary across tables. Future work will include external validation and testing on real hospital environments to assess clinical feasibility and deployment challenges.