Study of Hybridized Support Vector Regression Based Flood Susceptibility Mapping for Bangladesh

Flooding has become an exceedingly complex problem in many developing countries of the world including Bangladesh. Currently, Bangladesh is using MIKE 11 hydrodynamic model for flood forecasting. Previous studies indicated that hybridized machine learning models, especially support vector regression (SVR) models outperform standalone machine learning and other numerical models in mapping flood susceptibility. However, no study has been conducted on the flood dataset of Bangladesh using hybridized SVR model to predict flood susceptibility. In the present study, we have collected and modeled the recent flood inundation dataset of Bangladesh in terms of nine flood factors and explored their relative importance rank using the random forest (RF) algorithm. Then, we employed a genetic algorithm (GA) optimized SVR with radial basis function (RBF) kernel (hybridized GA-RBF-SVR) model along with the stand-alone RBF-SVR and multilayer perceptron (MLP) models to predict the flood susceptibility map for the whole country. The result of the hybridized SVR model is very promising to be employed in decision making to deal with the flood forecasting problem in Bangladesh.