• Six hybrid machine learning models are proposed to evaluate flood susceptibility. • NGBoost captures predictive uncertainty in flood susceptibility mapping • A novel meta-heuristic algorithm designed for machine learning is introduced. • Framework enhances reliability of spatial flood susceptibility models Floods represent a hazardous natural disaster, and their frequency is anticipated to rise because of climate change. The first step to create strategies for mitigation and minimize damage is to identify the distribution of flood susceptibility (FS), throughout a region. FS refers to the probability of flooding occurring in a particular area and is relies on the analysis of past data. involves a dual process of using a ML Algorithm combined with an Optimisation Algorithms to enhance the accuracy, efficiency, and scalability of predictions. This study proposes several novel hybrid models to determine FS, based on the optimization of several ML algorithms random forest (RF) and natural gradient boost (NGBoost) using different meta-heuristic algorithms genetic algorithm (GA), Dingo optimization algorithm (DOA) and differential evolution (DE). The suggested hybrid models were developed using flood inventory maps and twelve conditioning factors for training. The receiver operator characteristics curve (ROC) and statistical metrics were used to assess the validation of the models. It was demonstrated that all models performed strongly, achieving an area under the curve (AUC) value > 0.93. Among them, the RF-DOA model emerged as the most accurate, while the RF-DE model also showed strong performance. These findings suggest that the models are quite reliable in forecasting FS, though there are some minor differences in how they perform on certain measures. However, the RF-GA model proved to be the highest performer in terms of computational efficiency. The analysis revealed that river density has the most significant influence across all models in forecasting FS. This study highlighted that the six optimised models were highly effective in identifying regions at risk of flooding, which is of great help to emergency management decision makers. As such, the prepared FS map can serve as a valuable tool for informing strategies for reducing flood hazard risk.
Hassanvand et al. (Wed,) studied this question.