Investigation on the Association Between Socio-Economic Multivariate Data and Fire Incidence Based on Machine Learning Method: A Case Study in Shaanxi, China

The vast majority of fire incidents are caused by human factors, leading to a consensus that the occurrence of individual fire incident is accidental, while many scholars believe that there is a certain statistical pattern behind the randomness of a large number of fire incidents. In this paper, the fire records in Shaanxi Province from 2010 to 2020 are utilized to construct a panel data model based on which two machine learning models including Random Forest (RF) model and Back Propagation Neural Network (BPNN) model are trained and used to analyze the relationship between socio-economic multivariate factors and fire incidence/fire risk level. The fire incidence is predicted based on regression analysis based on machine learning models, while the fire risk level is predicted based on the classification capability of machine learning models. The results show that both the optimal RF model and BPNN model perform well in the regression task, with R² values to be 0.81 and 0.67, respectively; and can perform better in the prediction of fire risk level, with accuracy to be 91% and 91%, respectively. The results also reveal that socio-economic factors such as the population factors and economic factors could have the greatest importance for fire incidence prediction. This study demonstrates a significant correlation between fire incidence and socio-economic multivariate data, and also provides an important reference for regional fire risk assessment and fire incident prevention.