Can Deep Learning Algorithms Outperform Benchmark Machine Learning Algorithms in Flood Susceptibility Modeling

Abstract This paper introduces a new deep-learning algorithm of deep belief network (DBN) based on an extreme learning machine (ELM) that is structured by back propagation (BN) and optimized by particle swarm optimization (PSO) algorithm, named DEBP, for flood susceptibility mapping in the Vu Gia-Thu Bon watershed, central Vietnam. We use 847 locations of floods that occurred in 2007, 2009, and 2013 and 16 flood conditioning factors evaluated by an information gain ratio (IGR) technique to construct and validate the proposed model. Statistical metrics, including sensitivity, specificity, accuracy, F1-measure, Jaccard coefficient, Matthews correlation coefficient (MCC), root mean square error (RMSE), and area under the receiver operating characteristic curve (AUC), are used to assess the goodness-of-fit/performance and prediction accuracy of the new deep learning model. We further compare the proposed model with several well-known machine learning algorithms, including artificial neural network-based radial base function (ANNRBF), logistic regression (LR), logistic model tree (LMTree), functional tree (FTree), and alternating decision tree (ADTree). The new proposed model, DEBP, has the highest goodness-of-fit (AUC=0.970) and prediction accuracy (AUC=0.967) of all of the tested models and thus shows promise as a tool for flood susceptibility modeling. We conclude that novel deep learning algorithms such as the one used in this study can improve the accuracy of flood susceptibility maps that are required by planners, decision makers, and government agencies to manage of areas vulnerable to flood-induced damage.