Developing a disaster risk index for coastal communities in southwest Bangladesh: Shifting from data-driven models to holistic approaches

Communities in disaster-prone coastal areas are more vulnerable to socioeconomic, environmental, and ecological problems. Knowing how disaster-prone coastal communities are, and, ideally, their resilience against current disaster risks, is an essential first step in reducing consequences. Therefore, this study aims to develop a disaster risk index (DRI) for assessing disaster risks in three coastal communities in southwest Bangladesh. Five machine learning (ML) models and holistic approaches were adopted to conduct the analysis. This study utilized ML-based feature selection methods and considered hazard, exposure, and vulnerability components to identify the most influential indicators for evaluating disaster risk. The researchers systematically developed six distinct feature sets using filter-based feature selection methods. Five ML models then evaluate these sets to determine the optimal set of features. Evaluation metrics were used to validate the performance of ML models, such as accuracy, precision, F-measure, recall, and the receiver operating characteristics (ROC) curve. Based on the Logit Boost ML model, the optimal feature set (set 3) had the best accuracy (0.997), precision (0.997), recall (0.997), Fmeasure (0.997), and ROC (0.996) compared to other feature selection sets. According to the findings, Gabura had the highest DRI value of 0.09, ahead of the Southkhali (0.06) and Banishanta (0.04) unions. The study underscores the importance of evolving from traditional approaches to advanced data-driven tools to understand disaster risks. The paper's findings can provide policymakers in coastal areas with new insight into the disaster risk differences among the inhabitants, leading to more effective disaster management practices and risk reduction strategies.