Climate change is exacerbating the risks of floods a more significant threat to global stability and ecosystems, particularly in vulnerable regions like the Teesta River Basin in India with an estimated flood risk rate of 85%. The study generates comprehensive flood susceptibility maps by combining other important geospatial elements with digital elevation model (DEM) derived from Cartosat-1 satellite data. This study employed a novel approach implementing strategic data augmentation and Boruta analysis to optimize the selection of flood-conditioning factors, with advanced machine learning techniques- K-nearest neighbors (KNN), Naïve Bayes (NB), and decision trees (DT), ensuring the robustness of the predictive models for the study area. To determine flood-prone area prediction, these models were assessed using accuracy, precision, recall, F1-score, and receiver operating characteristics area under the curve (ROC-AUC). The findings show that the KNN model outperforms the NB and DT models in terms of accuracy of 91.62% and demonstrating robustness with ROC-AUC of 97.80%, 94.20%, 95.00% in training, testing and validation respectively, effectively delineating areas with varying levels of flood risk. This research provides flood risk management insights and a methodological framework for flood preparedness and resilience in flood-prone regions.
Authors: Deepanjan Sen, Swarup Das
DOI: https://doi.org/10.1201/9781003663348-47
Publish Year: 2025
