Hybrid AI Models for Continuous Vital Sign Monitoring: A Comparative Study of Deep Learning and Statistical Approaches

Abstract

Continuous monitoring of vital signs has emerged as a cornerstone of modern telehealth and remote patient-care systems. The proliferation of wearable sensors, mobile health applications, and IoT-enabled clinical instruments has created vast volumes of physiological data that demand highly efficient and interpretable predictive models. This study evaluates hybrid Artificial Intelligence (AI) models that integrate deep learning architectures (CNN, LSTM, GRU, Autoencoders) with classical statistical frameworks (ARIMA, Logistic Regression, Naïve Bayes, SVM) to enhance the accuracy, stability, and generalizability of continuous vital sign monitoring systems. The research investigates hybrid fusion strategies, including model stacking, weighted averaging, and parallel ensemble learning. Data were collected from a multi-source clinical dataset containing 24-hour readings of heart rate, blood pressure, respiratory rate, oxygen saturation, skin temperature, and ECG waveform segments. The study aims to determine how hybrid models outperform single-model approaches in predictive accuracy, anomaly detection, and early deterioration forecasting. The findings demonstrate that hybrid systems significantly improve the precision of sudden physiological change predictions, reduce false alerts, and enhance long-term stability. These outcomes lay the foundation for real-time clinical decision support systems within telemonitoring platforms.