Evaluation of landslide susceptibility based on SMOTE-Tomek sampling and machine learning algorithm.

Area of Science:

• Geosciences
• Geological Engineering
• Environmental Science

Background:

• Landslides are significant geological hazards impacting safety and infrastructure.
• Current landslide susceptibility assessments often lack sufficient data and rely on subjective experience.
• Accurate assessments are vital for effective risk management and mitigation strategies.

Purpose of the Study:

• To develop and validate a robust framework for assessing landslide susceptibility at the township scale.
• To compare the performance of multiple machine learning models for landslide susceptibility mapping.
• To identify key influencing factors on landslide occurrence using explainable AI.

Main Methods:

• Utilized a dataset of 1,325 slope units with nine characteristics in Taiping Township, China.
• Applied data balancing techniques, including Synthetic Minority Oversampling Technique and Tomek link (SMOTE-Tomek).
• Compared six machine learning models and employed the SHapley Additive exPlanation (SHAP) for factor analysis.

Main Results:

• The random forest (RF) model demonstrated optimal accuracy (0.791) and F1-score (0.723).
• Identified very low, low, medium, and high sensitivity zones covering 92.27%, 5.12%, 1.78%, and 0.83% of the area, respectively.
• Determined that cut slope height is the most significant factor, while altitude has a minor influence.

Conclusions:

• The proposed machine learning framework accurately assesses landslide susceptibility at the township scale.
• The findings provide valuable data for targeted risk management and mitigation efforts.
• Highlight the importance of considering specific geomorphological factors like cut slope height in landslide prediction.