A Resource-Efficient Boosting Ensemble Architecture for Continuous Gestational Diabetes Risk Assessment via Wearable IoT Bio-Sensors and Mobile Edge Computing

• Goodman Aze

  Author

Gestational Diabetes Mellitus (GDM) presents a significant global health challenge, with its prevalence rising steadily and posing substantial risks to both maternal and fetal well-being. Current diagnostic paradigms, reliant on the Oral Glucose Tolerance Test typically administered between 24 and 28 weeks of gestation, are reactive, failing to enable early intervention. While wearable Internet of Things (IoT) bio-sensors and machine learning offer promising avenues for continuous health monitoring, existing predictive models are often computationally intensive, unsuitable for resource-constrained edge devices, and lack validated architectures for continuous GDM risk assessment. This study proposes a novel resource-efficient boosting ensemble architecture, deployed at the mobile edge, for the continuous and interpretable risk assessment of GDM using data from wearable IoT bio-sensors. The methodology integrates a Light Gradient Boosting Machine (LightGBM) as the core ensemble learner, optimized for low-latency inference and minimal memory footprint, with a stream-based data processing pipeline for real-time feature extraction from physiological signals. The framework was validated using a simulated dataset representing continuous sensor streams from a cohort of 500 pregnant individuals. The proposed architecture achieved a predictive accuracy of 89.4% in identifying high-risk GDM cases, outperforming baseline models such as AdaBoost (84.2%) and Gradient Boosting (86.1%) while significantly reducing inference time. This architecture offers a practical, scalable, and resource-aware solution for integrating continuous GDM risk assessment into routine antenatal care via mobile health platforms.

• PDF

Copyright (c) 2026 Goodman Aze (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.