Comparative Analysis of Risk Factor Weighting GBDT Methods for Enhancing the Accuracy of Flood Risk Assessment

In recent years, the increasing frequency of flood disasters has emerged as a significant challenge to sustainable urban development. To improve the precision of urban flood risk assessment and effectively identify flood-prone cities, this study proposes an enhanced evaluation framework that integrates the traditional entropy weighted method with a genetic algorithm (GA) and gradient boosting decision tree (GBDT) for a weighted estimation of risk factors and indicators in flood risk assessments. Utilizing historical data from 16 prefecture-level cities in Shandong Province, China, the model assigns optimized weights to four key dimensions of flood risk or risk factors (hazard, exposure, vulnerability, and tolerance) and constructs a comprehensive urban flood risk assessment model. The resulting flood risk indices provide a quantitative basis for ranking the flood susceptibility of these cities, offering critical insights into targeted disaster risk reduction strategies. Subsequently, the assessment results were validated against urban flood loss data from 2017 to 2019. The GA, as a robust global optimization technique, has been extensively applied to identify optimal solutions for parameterized functions. The GBDT algorithm operates by minimizing the root mean squared error during iterative learning processes, incrementally refining the regression outcomes of the loss function to approach the optimal prediction. Comparative analysis demonstrates that relative to traditional approaches and the GA-based method, the GBDT model achieves substantially higher accuracy in identifying urban flood susceptibility, with error reductions of 8.94%, 24.23%, and 26.05%, respectively. These findings indicate that the proposed flood risk index serves as a reliable and effective decision-support tool, enabling policymakers to rapidly identify cities with inadequate disaster preparedness.