Improving flood-prone areas mapping using geospatial artificial intelligence (GeoAI): A non-parametric algorithm enhanced by math-based metaheuristic algorithms.
Journal:
Journal of environmental management
PMID:
39874699
Abstract
Flooding presents substantial dangers to human lives and infrastructure, underscoring the need to map flood-prone areas to implement effective mitigation measures precisely. Although machine learning algorithms have made great strides, their accuracy in flood susceptibility mapping (FSM) remains limited due to data dependence, interpretability, and explainability issues, overfitting, generalization difficulties, and hyperparameter tuning. This study suggests combining the Decision Tree (DT) algorithm with advanced, math-based metaheuristic optimization algorithms to address these limitations. In this study, the FSM was prepared using a combination of the DT non-parametric algorithm and three math-based metaheuristic optimization algorithms, including the gradient-based optimizer (GBO), chaos game optimization (CGO), and arithmetic optimization algorithm (AOA). Four models were developed and compared for modeling and mapping: the basic DT, DT-AOA, DT-GBO, and DT-CGO. The DT-CGO model stood out on the training set, achieving the lowest root mean square error (RMSE) of 0.17. It was closely followed by DT-AOA (0.19), DT-GBO (0.2), and DT (0.201). In terms of mean absolute error (MAE), the DT-CGO model exhibited the lowest value of 0.06, followed by DT-AOA (0.07), DT-GBO (0.079), and the basic DT model (0.08). When considering the coefficient of determination (R), the DT-CGO model achieved the highest value of 0.871, followed by DT-AOA (0.853), DT-GBO (0.84), and DT (0.838). The enhanced models performed superior over the test set's basic DT model. The area under the curve (AUC) values validated the enhanced models' efficacy, with the DT-CGO model achieving the highest AUC of 0.978, followed by DT-AOA (0.974), DT-GBO (0.967), and DT (0.958). The findings emphasize the importance of accurate flood-prone area identification and provide valuable insights for policymakers to develop robust flood mitigation strategies.