Citation

Hong H, Liu J, Zhu AX. Sci. Total Environ. 2020; 718: e137231.

Affiliation

Key Laboratory of Virtual Geographic Environment (Nanjing Normal University), Ministry of Education, Nanjing 210023, China; State Key Laboratory Cultivation Base of Geographical Environment Evolution (Jiangsu Province), Nanjing 210023, China; Jiangsu Center for Collaborative Innovation in Geographic Information Resource Development and Application, Nanjing, Jiangsu 210023, China; State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China; Department of Geography, University of Wisconsin-Madison, Madison, WI 53706, USA; Center for Social Sciences, Southern University of Science and Technology, Shenzhen, Guangzhou 518055, China.

Copyright

(Copyright © 2020, Elsevier Publishing)

DOI

10.1016/j.scitotenv.2020.137231

PMID

32097835

Abstract

The major target of this study is to design two novel hybrid integration artificial intelligent models, which are denoted as LADT-Bagging and FPA-Bagging, for modeling landslide susceptibility in the Youfanggou district (China). First of all, we prepared a geospatial database in the study area, including 79 landslide points that were divided into a training and validating dataset and 14 landslide conditioning factors. Second, the Support Vector Machines classifier (SVMC) approach was adapted to analyze the predictive capability of the landslide predisposing factors in each method. Then, a multicollinearity analysis using TOL and VIF parameters and Pearson's correlation coefficient methods were applied to verify the multicollinearity and correlation between these factors. Third, the LADT-Bagging and FPA-Bagging models were built by the integration of the LogitBoost alternating decision trees (LADT) with the Bagging ensemble and Forest by Penalizing Attributes (FPA) with the Bagging ensemble, respectively. Besides, heuristic tests were also applied to identify the appropriate values of each model's parameters in order to obtain the best programmer. Finally, for the training dataset, the results reveal that the LADT-Bagging model acquire the largest AUC value (0.980), smallest standard error (SE) (0.0134), narrowest 95% confidence interval (CI) (0.920-0.999), highest accuracy value (AV) (91.03%), highest specificity (94.44%), highest sensitivity (88.10%), highest F-measure (0.9115), lowest MAE (0.2016), lowest RMSE (0.2653), and highest Kappa (0.8205). About the result of validating dataset, it reveal that the LADT-Bagging model acquire the largest AUC value (0.781), the smallest SE (0.0539), the narrowest 95% CI (0.673-0.867), highest AV (71.19%), highest specificity (74.29%), highest sensitivity (69.77%), highest F-measure (0.7195), lowest MAE (0.3509), lowest RMSE (0.4335), and highest Kappa (0.4359). The results indicate that the LADT-Bagging model outperforms the FPA-Bagging, LADT and FPA models. Furthermore, the results of a Wilcoxon signed-rank test demonstrate that LADT-Bagging is significantly statistically different from other models. Therefore, in this study, the proposed new models are useful tools for land use planners or governments in high landslide risk areas.

Copyright © 2020 Elsevier B.V. All rights reserved.

Language: en

Keywords

Bagging; Forest by penalizing attributes; Integration model; Landslide; LogitBoost alternating decision trees