Data-Driven Ergonomic Risk Analysis Using Wearable Sensor Networks and Deep Learning for Injury Prevention in Industrial Workplaces

Abstract

This study investigated a data-driven ergonomic risk analysis framework integrating wearable sensor networks and deep learning to improve injury prevention in industrial workplaces. A total of 112 production workers participated after data screening, representing assembly, material handling, and mixed-task roles. Continuous biomechanical data, including trunk deviation duration, repetition rate, upper-limb elevation frequency, and muscular activation intensity, were collected across full work shifts. Multiple regression analysis revealed that wearable-derived exposure variables significantly predicted ergonomic risk scores, with the overall model explaining 51.7% of variance (R² = 0.517, p < .001). Trunk deviation duration (β = 0.41, p < .001) and repetition rate (β = 0.33, p < .001) emerged as the strongest predictors of elevated risk classification. The deep learning model achieved a classification accuracy of 84.6%, outperforming logistic regression (73.2%) and decision tree models (69.8%), with statistically significant differences (p < .001). Cumulative exposure analysis indicated a positive correlation between aggregated risk index and high-risk event frequency (r = 0.50, p < .001), confirming the influence of sustained biomechanical load. Reliability testing demonstrated strong internal consistency for the ergonomic exposure construct (α = 0.87) and cumulative risk index (α = 0.82). Material handling tasks accumulated the highest proportion of high-risk windows at 30.9%, compared to 14.0% in repetitive assembly operations. Findings demonstrated that wearable sensor–based predictive modeling provided reliable, objective, and generalizable ergonomic risk assessment, supporting its application as a scalable injury prevention strategy in industrial production systems.