TL;DR:
- Occupational hazards in the Korean construction sector are on the rise, with the highest number of accidents and fatalities reported in 2021.
- Korea’s Occupational Safety and Health Agency has introduced VR-based construction safety training as part of educational initiatives.
- Current VR-based training methods are passive and lack objective evaluation processes.
- Researchers at Incheon National University propose a machine learning approach using real-time biometric responses to predict personal learning performance during VR-based safety training.
- The study involved 30 construction workers, combining biometric responses, pre-training surveys, and post-training tests to develop forecasting models.
- Two models were developed: a full forecast model (FM) and a simplified forecast model (SM), with the SM being the more practical choice due to reduced complexity and overfitting.
- Dr. Choongwan Koo emphasizes the potential for improved safety and a secure working environment with this approach.
- The need for future research to consider various accident types and hazard factors in VR-based safety training is highlighted.
Main AI News:
In South Korea, the construction sector faces a mounting challenge: a surge in occupational hazards. Disturbingly, the Ministry of Employment and Labor’s ‘Occupational Safety Accident Status’ report for 2021 revealed that this industry accounted for the highest number of accidents and fatalities. In response, the Korea Occupational Safety and Health Agency has embarked on an innovative journey, integrating virtual reality (VR) technology into construction safety training, targeting daily workers as part of their educational initiatives.
Nonetheless, the current landscape of VR-based training methods grapples with two significant limitations. Firstly, VR-based construction safety training remains primarily passive, offering one-way instructions that fail to adapt to learners’ judgments and decisions. Secondly, the absence of an objective evaluation process during VR-based safety training poses a challenge. To confront these hurdles head-on, researchers have introduced immersive VR-based construction safety content to foster active worker engagement and have implemented post-written tests. Yet, these tests fall short in terms of immediacy and objectivity. Moreover, cognitive characteristics, one of several factors influencing learning performance, may undergo changes during VR-based safety training.
To tackle these issues, a dedicated team of researchers, led by Associate Professor Choongwan Koo from Incheon National University’s Division of Architecture & Urban Division, presents an innovative machine learning approach. Their goal? Predicting personal learning performance in VR-based construction safety training using real-time biometric responses. Their groundbreaking research was published online on October 7, 2023, and is set to appear in Volume 156 of the journal “Automation in Construction” this December.
Dr. Koo explains the significance of their approach: “Traditional methods relying on post-written tests may lack objectivity. Real-time biometric responses, acquired through eye-tracking and electroencephalogram (EEG) sensors, can promptly and objectively evaluate personal learning performance during VR-based safety training.”
Their study involved 30 construction workers engaged in VR-based construction safety training. Real-time biometric responses, including eye-tracking and EEG data to monitor brain activity, were meticulously collected during the training to assess participants’ psychological responses. By combining this data with pre-training surveys and post-training written tests, the researchers crafted machine-learning-based forecasting models to gauge the overall personal learning performance of participants during VR-based safety training.
The team developed two models: a full forecast model (FM), utilizing both demographic factors and biometric responses as independent variables, and a simplified forecast model (SM), which relied solely on principal features as independent variables to reduce complexity. Although the FM exhibited superior accuracy in predicting personal learning performance compared to traditional models, it also exhibited a high degree of overfitting. In contrast, the SM displayed higher prediction accuracy than the FM due to its reduced number of variables, significantly mitigating overfitting. Consequently, the researchers concluded that the SM was the most practical choice.
Dr. Koo underscores the potential impact of their approach, stating, “This method has the potential to substantially enhance personal learning performance during VR-based construction safety training, ultimately preventing safety incidents and fostering a secure working environment.” Additionally, the team calls for future research to delve into various accident types and hazard factors in VR-based safety training, further solidifying the foundations of a safer tomorrow.
Conclusion:
The introduction of a real-time biometric response prediction model for VR-based safety training holds significant promise for the market. It has the potential to revolutionize the way safety training is conducted, improving personal learning outcomes, reducing incidents, and fostering safer working environments, thus enhancing the market’s safety training landscape.
