首页|Landslide susceptibility mapping using J48 Decision Tree with AdaBoost, Bagging and Rotation Forest ensembles in the Guangchang area (China)

Landslide susceptibility mapping using J48 Decision Tree with AdaBoost, Bagging and Rotation Forest ensembles in the Guangchang area (China)

扫码查看
原文链接
  • NSTL
  • Elsevier
Landslides are a manifestation of slope instability causing different kinds of damage affecting life and property. Therefore, high-performance-based landslide prediction models are useful to government institutions for developing strategies for landslide hazard prevention and mitigation. Development of data mining based algorithms shows that high-performance models can be obtained using ensemble frameworks. The primary objective of this study is to investigate and compare the use of current state-of-the-art ensemble techniques, such as AdaBoost, Bagging, and Rotation Forest, for landslide susceptibility assessment with the base classifier of J48 Decision Tree (JDT). The Guangchang district (Jiangxi province, China) was selected as the case study. Firstly, a landslide inventory map with 237 landslide locations was constructed; the landslide locations were then randomly divided into a ratio of 70/30 for the training and validating models. Secondly, fifteen landslide conditioning factors were prepared, such as slope, aspect, altitude, topographic wetness index (TWI), stream power index (SPI), sediment transport index (STI), plan curvature, profile curvature, lithology, distance to faults, distance to rivers, distance to roads, land use, normalized difference vegetation index (NDVI), and rainfall. Relief-F with the 10-fold cross-validation method was applied to quantify the predictive ability of the conditioning factors and for feature selection. Using the JDT and its three ensemble techniques, a total of four landslide susceptibility models were constructed. Finally, the overall performance of the resulting models was assessed and compared using area under the receiver operating characteristic (ROC) curve (AUC) and statistical indexes. The result showed that all landslide models have high performance (AUC > 0.8). However, the JDT with the Rotation Forest model presents the highest prediction capability (AUC = 0.855), followed by the JDT with the AdaBoost (0.850), the Bagging (0.839), and the JDT (0.814), respectively. Therefore, the result demonstrates that the JDT with Rotation Forest is the best optimized model in this study and it can be considered as a promising method for landslide susceptibility mapping in similar cases for better accuracy.

J48 Decision TreeAdaBoostBaggingRotation ForestGuangchangGIS

Liu, Junzhi、Dieu Tien Bui、Pradhan, Biswajeet、Acharya, Tri Dev、Hong, Haoyuan、Binh Thai Pham、Zhu, A-Xing、Chen, Wei、Bin Ahmad, Baharin

展开 >

Nanjing Normal Univ, Minist Educ, Key Lab Virtual Geog Environm, Nanjing 210023, Jiangsu, Peoples R China

Telemark Univ Coll, Dept Econ & Comp Sci, Geog Informat Syst Grp, N-3800 Bo I Telemark, Norway

Univ Technol Sydney, Fac Engn & IT, Sch Syst Management & Leadership, CB11 06 217,Bldg 11,81 Broadway,POB 123, Ultimo, NSW 2007, Australia

Kangwon Natl Univ, Dept Civil Engn, Chunchon, South Korea

Gujarat Technol Univ, Dept Civil Engn, Visat Gandhinagar Highway, Ahmadabad 382424, Gujarat, India

Xian Univ Sci & Technol, Coll Geol & Environm, Xian 710054, Shaanxi, Peoples R China

Univ Teknol Malaysia, Fac Geoinformat & Real Estate, Dept Geoinformat, Skudai, Malaysia

展开 >

2018

10.1016/j.catena.2018.01.005

Catena

Catena

ISTP
ISSN：0341-8162
年,卷(期)：2018.163
  • 177
  • 89