Harnessing Machine Learning and Multi-source Data Fusion for Urban Fire Risk Assessment: Predictive Analysis of Spatial Heterogeneity

As a giant system with many functional elements and complex structural layout, the occurrence of fire in megacities is influenced by multiple factors such as building type, regional planning, and socio-economic environment, resulting in a high uncertainty in the probability and risk of fire occurring in different areas of the city. This paper quantitatively analyzed the factors influencing urban fire risk by combining historical fire event data and the development characteristics of megacities. Based on the key features of fire risk domain knowledge, a model for Urban Area Fire Probability Prediction (UAFPP) was constructed using machine learning and multi-source data fusion technology. The quantitative relationship between the regional fire probability and the economic population, social activities, urban land use type was analyzed. The UAFPP model exhibits robust predictive performance for urban fires, evidenced by consistent R2 and RMSE values across a three-year real fire accident data validation period. This confirms the effectiveness of capturing the spatial characteristics of urban fire risk within the framework. Through a Shenzhen case study, we analyzed spatial patterns and predicted metropolitan fire probabilities. We quantified risk characteristics using multi-source data to provide actionable insights for mitigating urban fire risks.