Scholarlink

<p>Urban fires increasingly threaten the resilience of modern cities, particularly in densely populated areas with aging infrastructure. Traditional post-fire damage assessments, which rely on manual inspections and expert evaluations, are often time-consuming, subjective, and infeasible during large-scale emergencies. This study proposes an interpretable machine learning framework for rapid prediction of fire-induced structural damage and prioritization of building rehabilitation. A comprehensive dataset of 500 fire-affected buildings was compiled, incorporating structural features, fire exposure metrics, environmental conditions, and occupancy-related variables. Seven supervised learning algorithms—MLR, SVR, RFR, GBM, ANN, XGBoost, and CatBoost—were trained to predict a normalized Structural Damage Index (SDI). CatBoost emerged as the best-performing model with an R² of 0.89 and RMSE of 0.071. SHAP-based analysis identified maximum temperature, fire duration, retrofitting status, and population density as the most influential predictors. To translate predictions into actionable insights, a Rehabilitation Priority Score (RPS) was developed by integrating SDI with contextual factors such as population exposure, infrastructure criticality, and accessibility. The RPS rankings strongly aligned with expert judgment (Spearman’s ρ &gt; 0.85), enabling equitable and transparent post-disaster recovery planning. The proposed framework supports scalable, data-driven decision-making and is well-suited for integration with GIS platforms to enhance urban fire resilience.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Integrating machine learning algorithms for post-fire damage assessment and rehabilitation prioritization of urban structures

  • 期刊论文

  • 发表日期：

    2026年1月22日

查看全文
  • Dileep Kumar M.,
  • Mayank Chauhan,
  • Kamal Sharma,
  • Barada Prasad Sethy,
  • Gohar Ali,
  • Krushna Chandra Sethi

摘要

Urban fires increasingly threaten the resilience of modern cities, particularly in densely populated areas with aging infrastructure. Traditional post-fire damage assessments, which rely on manual inspections and expert evaluations, are often time-consuming, subjective, and infeasible during large-scale emergencies. This study proposes an interpretable machine learning framework for rapid prediction of fire-induced structural damage and prioritization of building rehabilitation. A comprehensive dataset of 500 fire-affected buildings was compiled, incorporating structural features, fire exposure metrics, environmental conditions, and occupancy-related variables. Seven supervised learning algorithms—MLR, SVR, RFR, GBM, ANN, XGBoost, and CatBoost—were trained to predict a normalized Structural Damage Index (SDI). CatBoost emerged as the best-performing model with an R² of 0.89 and RMSE of 0.071. SHAP-based analysis identified maximum temperature, fire duration, retrofitting status, and population density as the most influential predictors. To translate predictions into actionable insights, a Rehabilitation Priority Score (RPS) was developed by integrating SDI with contextual factors such as population exposure, infrastructure criticality, and accessibility. The RPS rankings strongly aligned with expert judgment (Spearman’s ρ > 0.85), enabling equitable and transparent post-disaster recovery planning. The proposed framework supports scalable, data-driven decision-making and is well-suited for integration with GIS platforms to enhance urban fire resilience.