Shallow Landslide Susceptibility Mapping: A Comparison between Logistic Model Tree, Logistic Regression, Naïve Bayes...

Area of Science:

• Geosciences
• Environmental Science
• Data Science

Background:

• Shallow landslides pose significant risks to infrastructure, agriculture, and human life.
• Landslide susceptibility mapping is crucial for effective land-use management and hazard mitigation.

Purpose of the Study:

• To compare the effectiveness of five machine learning algorithms for shallow landslide susceptibility mapping.
• To identify the best-performing model for reliable landslide hazard assessment in Bijar City, Iran.

Main Methods:

• Five machine learning algorithms were evaluated: Logistic Model Tree (LMT), Logistic Regression (LR), Naïve Bayes Tree (NBT), Artificial Neural Network (ANN), and Support Vector Machine (SVM).
• Twenty conditioning factors were used to model 111 shallow landslides.
• The One-R attribute evaluation (ORAE) technique was employed for modeling and validation.
• Model performance was assessed using statistical indexes like accuracy, MAE, RMSE, and AUC.

Main Results:

• All five machine learning models demonstrated good performance in shallow landslide susceptibility assessment.
• The Logistic Model Tree (AUC = 0.932) and Logistic Regression (AUC = 0.932) models exhibited the highest goodness-of-fit and prediction accuracy.
• Naïve Bayes Tree (AUC = 0.864), ANN (AUC = 0.860), and SVM (AUC = 0.834) also showed strong predictive capabilities.

Conclusions:

• The Logistic Model Tree model is recommended for shallow landslide mapping in semi-arid regions due to its superior performance.
• Findings support decision-makers, planners, and agencies in mitigating landslide hazards and risks.