Optimizing machine learning and bagging-based hybrid models for landslide susceptibility mapping: a case study in...

Area of Science:

• Geosciences
• Environmental Science
• Machine Learning Applications

Background:

• Landslide susceptibility mapping is crucial for hazard mitigation.
• Traditional methods often lack the predictive power of advanced computational models.
• Chenggu County faces significant landslide risks, necessitating accurate predictive tools.

Purpose of the Study:

• To compare the effectiveness of various models, including statistical and machine learning approaches, for landslide susceptibility mapping.
• To optimize machine learning models using hyper-parameter tuning for improved predictive accuracy.
• To identify the most effective model for landslide prediction in Chenggu County.

Main Methods:

• Utilized a Certainty Factor (CF) model and several machine learning algorithms: Functional Tree (FT), Logistic Model Tree (LMT), Alternating Decision Tree (ADT), and Reduced Error Pruning Tree (REPT).
• Developed hybrid models using bagging techniques (Bagging-FT, Bagging-ADT, Bagging-REPT, Bagging-LMT).
• Employed grid search for hyper-parameter optimization and Receiver Operating Characteristic (ROC) curves for accuracy assessment.

Main Results:

• The hyper-parameter optimized Bagging-Logistic Model Tree (Bag-LMT) model achieved the highest prediction accuracy.
• Comparative analysis using ROC curves validated the superior performance of the optimized hybrid model.
• The study identified 16 key landslide conditioning factors influencing landslide occurrence in the region.

Conclusions:

• Optimization of machine learning models significantly enhances geohazard prediction capabilities.
• The Bag-LMT model offers a robust and accurate tool for landslide susceptibility mapping in Chenggu County.
• Advanced modeling techniques provide valuable insights for effective land-use planning and disaster risk reduction.