Assessing urban fire risk: An ensemble learning approach based on scenarios and cases

Abstract

Urban fires represent a significant hazard to people’s lives and property, which makes it critical to estimate the risk adequately. Existing urban fire evaluation methods lack applicability because they do not take into account individual scene components and previous cases. As a result, this study offers the scenario- and case-based urban fire risk assessment approach (SCBUFRA), which seeks to achieve a more thorough and accurate urban fire risk assessment. First, the technique uses fire case and scenario data, as well as the recursive feature elimination method, to pick the elements utilized to assess urban fire risk. Second, the data-driven empowerment technique and stability analysis are utilized to determine the precise fire risk value and correctly quantify the fire danger level in each part of the city. Next, the Affinity Propagation (AP) technique is used to cluster scene elements. Ensemble learning is then used to create a risk prediction model by refining the weighting strategy of $R^2$. Finally, Shapley additive explanations are used to investigate the elements causing urban fires. The findings show that SCBUFRA outperforms popular machine learning methods, that the number of crimes, gross population, and house price are the most important variables for fire prediction, and that the research is applicable to urban fire risk management and firefighting resource allocation.