Title Machine Learning?Based Analysis of Construction Site Fire Risk Factors Considering Weather Conditions
Authors 김원창(Kim, Won-Chang) ; 최형길(Choi, Hyeong-gil) ; 이태규(Lee, Tae Gyu)
DOI https://doi.org/10.5659/JAIK.2026.42.8.365
Page pp.365-372
ISSN 2733-6247
Keywords Construction Site Fire; Machine Learning; Weather; Prediction Model
Abstract In this study, a machine learning-based multi-classification model was developed and evaluated to classify the risk of construction site fires using weather conditions. A database was created using the number of construction site fires provided by the Fire and Rescue Service and weather condition data provided by the Korea Meteorological Administration, and quartiles were used to set the criteria for construction site fire risk. Seven machine learning-based multi-classification algorithms were then used for modeling and evaluation. The cumulative weekly construction site fires were negatively correlated with weather conditions related to temperature (-0.84 to -0.75), negatively correlated with humidity (-0.83 to -0.793), and positively correlated with barometric pressure (0.78). We classified the construction site fire risk criteria into three levels (fire0, fire1, and fire2), and the XGBoost and LightGBM models showed the best performance in the modeling results. The statistical test of weather conditions by construction site fire risk showed that the higher the fire risk, the more winter and spring weather conditions, and there was a statistically significant difference between the groups at the 0.01 significance level.