Coupling RBF neural network with ensemble learning techniques for landslide susceptibility mapping

•Developing four ensemble models for landslide susceptibility mapping.•Testing ASC, CG, Dagging, Random Subspace techniques performance.•Reliable susceptibility mapping using the Decorate-RFBN ensemble.•Providing insights for developing more advanced landslide predictive models. Using multiple ensem...

Full description

Saved in:

Bibliographic Details
Published in:	Catena (Giessen) 2020-12, Vol.195, p.104805, Article 104805
Main Authors:	Pham, Binh Thai, Nguyen-Thoi, Trung, Qi, Chongchong, Phong, Tran Van, Dou, Jie, Ho, Lanh Si, Le, Hiep Van, Prakash, Indra
Format:	Article
Language:	eng
Subjects:	Ensemble modeling Landslide susceptibility mapping Machine learning Vietnam
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	•Developing four ensemble models for landslide susceptibility mapping.•Testing ASC, CG, Dagging, Random Subspace techniques performance.•Reliable susceptibility mapping using the Decorate-RFBN ensemble.•Providing insights for developing more advanced landslide predictive models. Using multiple ensemble learning techniques for improving the predictive accuracy of landslide models is an active research area. In this study, we combined a radial basis function (RBF) neural network (RBFN) with the Random Subspace (RSS), Attribute Selected Classifier (ASC), Cascade Generalization (CG), Dagging for spatial prediction of landslide susceptibility in the Van Chan district, Yen Yen Bai Province, Vietnam. A geospatial database that contained records from 167 historical landslides and 12 conditioning factors (slope, aspect, elevation, curvature, slope length, valley depth, topographic wetness index, and terrain ruggedness index, and distance to rivers, roads, and faults) were used to develop the ensemble models. The models were validated via area under the receiver operating characteristic curve (AUC) and several other performance metrics (i.e., positive predictive value, negative predictive value, sensitivity, specificity, accuracy, and Kappa). Although the single RBFN model (AUC = 0.799) performed better than the ensemble models (AUCaverage = 0.77) in the training phase, the ensemble models (AUCaverage = 0.83) outperformed RBFN (AUC = 0.79) in the validation phase, demonstrating superior predictive performance of the ensemble models for the prediction of future landslides. Our study provides insights for developing reliable landslide predictive models for different landslide-prone regions around the world.
ISSN:	0341-8162 1872-6887