Hierarchical Learning Models for Generalization Across Heterogeneous Data

Abstract

The rapid expansion of Internet of Things (IoT)-enabled telemonitoring systems has facilitated continuous healthcare monitoring through wearable sensors, smart medical devices, and remote patient monitoring platforms. However, conventional centralized learning approaches are constrained by privacy risks, communication overhead, scalability limitations, and performance degradation caused by Non-Independent and Identically Distributed (Non-IID) healthcare data across distributed devices. Existing federated and distributed learning frameworks often exhibit inefficient hierarchical coordination and limited adaptability in heterogeneous healthcare environments. To address these challenges, this research proposes a Hierarchical Deep Learning (HDL) model based on a Capuchin Search Algorithm-tuned Hierarchical Recurrent Neural Network (CSA-HRNN) for privacy-preserving telemonitoring applications. The model is evaluated using IoT-based telemonitoring healthcare datasets containing physiological signals, including heart rate, electrocardiogram (ECG), blood pressure, oxygen saturation, and body temperature measurements. Initially, the collected data are preprocessed using Min-Max normalization to eliminate scale variations and improve learning stability. Subsequently, Independent Component Analysis (ICA) is employed for feature extraction to obtain informative and statistically independent representations of physiological parameters. The proposed hierarchical architecture performs local model training on edge devices, followed by edge-level and cloud-level aggregation without sharing raw patient data. The model is implemented in Python. Experimental results demonstrate that the proposed CSA-HRNN model achieves an accuracy of 98.4%, precision of 0.983, recall of 0.982, and an F1-score of 0.981, while reducing communication overhead and accelerating convergence under Non-IID data environments. The findings confirm the effectiveness of the proposed model for reliable and intelligent IoT-based telemonitoring systems.