Modeling landslide susceptibility using LogitBoost alternating decision trees and forest by penalizing attributes...

Area of Science:

• Geosciences and environmental modeling.
• Artificial intelligence in natural hazard assessment.

Background:

• Landslide susceptibility mapping is crucial for risk management.
• Existing models require enhancement for improved accuracy.

Purpose of the Study:

• To design and evaluate two novel hybrid artificial intelligence models: LADT-Bagging and FPA-Bagging.
• To assess landslide susceptibility in the Youfanggou district, China.

Main Methods:

• Preparation of a geospatial database with landslide points and conditioning factors.
• Application of Support Vector Machines classifier (SVMC) for predictive analysis.
• Development of LADT-Bagging (LogitBoost alternating decision trees with Bagging) and FPA-Bagging (Forest by Penalizing Attributes with Bagging) models.
• Multicollinearity analysis using TOL, VIF, and Pearson's correlation coefficient.

Main Results:

• The LADT-Bagging model achieved superior performance on the training dataset with an AUC of 0.980 and high accuracy (91.03%).
• On the validating dataset, LADT-Bagging also outperformed other models with an AUC of 0.781 and accuracy of 71.19%.
• Statistical tests confirmed LADT-Bagging's significant superiority over FPA-Bagging, LADT, and FPA models.

Conclusions:

• The LADT-Bagging model is a highly effective tool for landslide susceptibility modeling.
• The proposed models can aid land use planners and governments in managing landslide risks.
• Hybrid AI integration offers a promising approach for enhancing natural hazard assessment.