Related Experiment Videos
Principled XAI analysis of the deep learning-based landslide susceptibility prediction model.
Jongchan Oh1, Jung-Hyun Lee2, Hyuck-Jin Park2
1Department of Energy Resources Engineering, Chonnam National University, Gwangju, 61186, Republic of Korea.
Scientific Reports
|June 14, 2026
Summary
Deep learning models, specifically Convolutional Neural Networks (CNNs), show superior performance in landslide susceptibility mapping compared to traditional machine learning methods. These advanced models improve prediction accuracy and reliability by effectively capturing spatial context.
Area of Science:
- Geosciences and Artificial Intelligence
- Machine Learning for Natural Hazard Assessment
Background:
- Machine learning (ML) and deep learning (DL) are increasingly used for landslide susceptibility analysis, offering high accuracy but often lacking interpretability.
- The 'black-box' nature of ML models raises concerns about the reliability of landslide susceptibility predictions.
- eXplainable Artificial Intelligence (XAI) methods are being explored to interpret and validate these complex models.
Purpose of the Study:
- To develop and compare landslide susceptibility prediction models using various ML and DL architectures.
- To evaluate the performance of traditional ML models against image-based DL models.
- To assess the interpretability and reliability of these models using XAI techniques.
Main Methods:
- Developed landslide susceptibility models using Support Vector Machine (SVM), Random Forest (RF), Multilayer Perceptron (MLP), and Convolutional Neural Networks (CNNs).
- Utilized 20 conditioning factors, including digital elevation models (DEM), forest characteristics, soil properties, and geological features.
- Applied XAI techniques, including SHapley Additive exPlanations (SHAP) and Gradient-weighted Class Activation Mapping (Grad-CAM) for model interpretation.
Main Results:
- Convolutional Neural Networks (CNNs) achieved the highest Accuracy (0.7586) and Recall (0.8138), significantly outperforming traditional models (RF, SVM, MLP).
- Image-wise input processing in CNNs proved more effective for landslide susceptibility mapping than pixel-level analysis.
- XAI methods, particularly Grad-CAM heatmaps for CNNs, effectively visualized patterns linking conditioning factors to landslide susceptibility, enhancing model interpretability.
Conclusions:
- Deep learning, especially CNNs, offers superior performance and reliability for landslide susceptibility mapping by leveraging spatial context.
- XAI techniques are crucial for understanding and validating the predictions of complex ML and DL models in geoscience applications.
- The findings highlight the potential of integrating advanced AI with geospatial data for improved natural disaster risk assessment.
