A deep-learning approach to leveraging natural hazard indicators for improved safety on construction sites

초록

The growing frequency of natural disasters and extreme weather events is posing increasingly critical safety hazards at construction sites. Existing research approaches using AI technologies underline safety risks themselves, but missing links with external factors causing safety accidents on sites. This study fills these this knowledge gaps by developing a deep-learning-driven predictive modeling framework that can accurately and reliably predict the most critical safety accident patterns on construction sites, the number of deaths, by incorporating both natural hazard indicators and building construction information into the modeling framework. A total of 36 different deep learning alternatives were modeled, compared, and validated scientifically. This study is the first of its kind, and the developed model contributes to predicting the most critical safety accidents caused by natural disaster factors. Main findings of this study are expected to help advanced safety management for construction projects coupled with natural disaster indicators.

키워드