Development of an Artificial Neural Network Model to Predict Fire Risk at Construction Sites Using Domestic Meteorological Data

This study developed a regression model based on an artificial neural network (ANN) to predict fire risk at construction sites based on domestic meteorological data. Monthly fire incident records and key meteorological variables from 2010 to 2023 were collected. Independent features were selected based on strong Pearson correlations, including temperature, humidity, and dew point. The performance comparison showed that the ANN model with a single hidden layer achieved the highest prediction accuracy, while additional hidden layers led to reduced performance and an increased risk of overfitting. Permutation importance analysis identified minimum temperature and apparent temperature as the most influential predictors.