Causal discovery and inference for urban fire

Based on the theoretical framework of Pearl's causal model, this study conducted research in three main areas to address the lack of work on causal model construction in the urban fire domain. First, at the association level, the PC (Peter-Clark) algorithm was used to construct an initial causal network model in the urban fire domain to establish causal relationships between relevant factors. However, due to potential missing directed relationships between casual network nodes, the network structure was supplemented based on knowledge provided by domain experts. Second, at the intervention level, the study applied the backdoor criterion and the front-door criterion to estimate the causal effects between fire variables in the above causal network, and determined the variable nodes with the greatest causal effect through three main variables and the final direction of the network. Finally, at the counterfactual level, the study tested and verified the constructed causal network structure using methods such as adding random confounding factors, placebo interventions, and data subset validation. The experimental results show that the proposed urban fire causal network obtained by the framework of causal analysis has a certain robustness and interpretability, demonstrating the effectiveness and reliability of the methods used.