A real time machine learning framework for multivital ailment detection in remote patient monitoring

As the number of wearable sensing and IoT-enabled healthcare systems continues to surge, remote patient monitoring is gaining attention for the early detection of abnormal physiological conditions. However, current methods are restricted to single-signal analysis, rely on intensive deep learning models, offer poor interpretability, or are not suitable for deployment on mobile and edge devices. In response to these limitations, this study proposes a framework named SmartHealthRPM+VitalTrackAI that is edge-oriented, addresses multi-vital ailment classification, and incorporates real-time ECG-derived rhythm descriptors, as well as blood pressure (BP), heart rate (HR), and SpO₂ measurements. The adapted VitalTrackAI-Gated fusion model incorporates modality-specific encoders, assesses signal quality, implements adaptive cross-vital gating, and gated multi-vital fusion to capture the dynamic dependencies among heterogeneous physiological signals. The four physiological conditions are: Normal, Arrhythmia, Hypertension, and Hypoxia, as defined by a compact classification head. To enhance the transparency of model decisions, the framework also includes confidence scoring and SHAP interpretability. Under the same evaluation protocol, the experimental evaluation demonstrates that the proposed model achieves 0.90 F1-score, 0.92 AUC, and 90% accuracy, outperforming the baselines of Logistic Regression, Random Forest, SVM, and MLP. Testing under noisy, missing, and fluctuating signal conditions reveals robust performance, and the edge test demonstrates low-latency inference, making it ready for deployment on smartphones. The proposed framework thus offers a practical, interpretable, and computationally efficient solution to real-time multi-vital patient monitoring in remote areas.