Improving Health Monitoring of Construction Workers Using Physiological Data-Driven Techniques: An Ensemble Learning-Based Framework to Address Distributional Shifts

While researchers have used various off-the-shelf physiological sensors and prevalent machine learning (ML) algorithms to objectively assess construction workers' health status, there remain specific challenges for consistent and accurate health monitoring on the jobsite. The existing physiological-based data-driven frameworks for predicting workers' health status in the field are not robust to the distribution shift of physiological signals and face challenges in stability, reliability, and accuracy. To overcome these issues, this paper proposes using an ensemble learning technique implemented on a support vector machine (SVM) with the Adaptive Boosting (AdaBoost) algorithm to develop a resilient predictive performance of the data- driven framework. To examine the performance of the framework, physiological signals were collected from 10 subjects performing material handling tasks with varying levels of physical fatigue. The proposed framework predicted the physical fatigue level with over 88% accuracy, better than single machine learning classifiers. This study has significant implications for improving the accuracy and stability of physiological-sensing-based health monitoring.