Automated fall risk classification for construction workers using wearable devices, BIM, and optimized hybrid deep learning

Abstract

With the highest rate of workplace fatalities, construction is one of the world's most hazardous industries. Current risk mitigation approaches, which still rely heavily on traditional methods, do not allow decision-makers to respond quickly and accurately to the dynamic changes that typify modern construction environments. To address this issue, this paper develops an automated worker fall risk monitoring system for dynamic construction sites, by integrating real-time data from wearable devices and BIM with optimized hybrid deep learning model. The model utilizes Neural Network (NN) for time-independent variables and Graph Neural Network (GNN) for time-dependent variables. Optimization is achieved through the Symbiotic Organisms Search (SOS), enhancing the model's architecture and output weights. The classification performance of SOS-NN-GNN consistently outperformed other models, which resulted in 90.98 % accuracy. This highlights the model's reliability in automatically detecting fall risk levels, significantly reducing fall-related accidents, and improving safety, efficiency, and project outcomes in construction engineering.