Flood susceptibility mapping using machine learning boosting algorithms techniques in Idukki district of Kerala India

Kerala experiences a high rate of annual rainfall and flooding resulting in a frequent natural disaster. The objective of this study is to develop flood susceptibility maps for the Idukki district making use of Remote Sensing (RS) data, Geographic Information System (GIS), and Machine Learning (ML). In this study, five different ML models (Adaboost, Gradient boosting, Extreme Gradient Boosting (XGB), CatBoost, Stochastic Gradient Boosting (SGB)) are evaluated to deter-mine flood susceptibility in Idukki district Kerala. There were a total of sixteen hydrometeoro-logical parameters taken into account. Area under the curve (AUC) was used to evaluate the accuracy of various techniques in terms of both prediction and success rates. The validation re-sults proved the efficiency of the individual techniques. The highest AUC was obtained by the SGB and GBC (92%), followed by that of the Adaboost with AUC 87%, and the lowest AUC was ob-tained by CatBoost, with AUC of 79%. The absence of data overfitting in all models demonstrates the efficacy of boosting techniques. The boosting algorithms penalize models that overfit the training set, which helps to decrease overfitting. Researchers and local governments could benefit from the proposed boosting techniques in the flood susceptibility mapping and mitigation strategies.

Automated summary

The objective of this study is to develop flood susceptibility maps for the Idukki district in Kerala using Remote Sensing (RS) data, Geographic Information System (GIS), and Machine Learning (ML). Five different ML models were evaluated and the results showed that SGB and GBC models had the highest AUC (92%), followed by Adaboost (87%) and CatBoost (79%). The absence of data overfitting in all models demonstrates the efficacy of boosting techniques in flood susceptibility mapping and mitigation strategies.