Assessment of construction workers’ perceived risk using physiological data from wearable sensors: A machine learning approach

Abstract

Considering that workers' safe or unsafe behaviors are responses to their perceived risk when working, understanding workers' perceived risk is vital for safety management in the construction industry. Existing tools for measuring workers' perceived levels of risk mainly rely on post-hoc survey-based assessments, which are limited by their lack of continuous monitoring ability, lack of objectivity, and high cost. To address these limitations, this study develops an automatic method to recognize construction workers’ perceived levels of risk by using physiological signals acquired from wristband-type wearable biosensors in conjunction with a supervised-learning algorithm. The performance of the model was examined with physiological signals acquired from eight construction workers performing their daily work. The model achieved a validation accuracy of 81.2% for distinguishing between low and high levels of perceived risk. This study provides a new means of continuous, objective, and non-invasive method for monitoring construction workers' perceived levels of risk.