| Title |
An interpretable machine learning approach for forecasting personal heat strain considering the cumulative effect of heat exposure |
| Authors |
Seo, Seungwon ; Choi, Yujin ; Koo, Choongwan |
| DOI |
https://dx.doi.org/10.6106/KJCEM.2023.24.6.081 |
| Keywords |
Personal Heat Strain; Cumulative Working Time; Machine Learning Algorithm; Forecast Model; Feature Importance |
| Abstract |
Climate change has resulted in increased frequency and intensity of heat waves, which poses a significant threat to the health and safety of construction workers, particularly those engaged in labor-intensive and heat-stress vulnerable working environments. To address this challenge, this study aimed to propose an interpretable machine learning approach for forecasting personal heat strain by considering the cumulative effect of heat exposure as a situational variable, which has not been taken into account in the existing approach. As a result, the proposed model, which incorporated the cumulative working time along with environmental and personal variables, was found to have superior forecast performance and explanatory power. Specifically, the proposed Multi-Layer Perceptron (MLP) model achieved a Mean Absolute Error (MAE) of 0.034 (℃) and an R-squared of 99.3% (0.933). Feature importance analysis revealed that the cumulative working time, as a situational variable, had the most significant impact on personal heat strain. These findings highlight the importance of systematic management of personal heat strain at construction sites by comprehensively considering the cumulative working time as a situational variable as well as environmental and personal variables. This study provided a valuable contribution to the construction industry by offering a reliable and accurate heat strain forecasting model, enhancing the health and safety of construction workers. |