Hybrid machine learning approach for landslide prediction, Uttarakhand, India

Area of Science:

• Geosciences and Environmental Science
• Artificial Intelligence and Machine Learning
• Data Science and Analytics

Background:

• Natural disasters like landslides cause significant global damage to human and natural resources.
• Accurate landslide prediction is crucial for mitigating risks and protecting vulnerable areas.
• Existing prediction models often require enhancement for improved accuracy and stability.

Purpose of the Study:

• To evaluate and compare the prediction accuracy of five hybrid models for landslide occurrence.
• To assess the effectiveness of Rough Set theory when combined with different machine learning models.
• To develop a user-friendly GIS platform for real-time landslide prediction.

Main Methods:

• Developed five hybrid models by coupling Rough Set theory with Bayesian Network (HBNRS), Backpropagation Neural Network (HBPNNRS), Bagging (HBRS), XGBoost (HXGBRS), and Random Forest (HRFRS).
• Utilized a database of 373 landslide and 181 non-landslide locations in Uttarkashi, Uttarakhand, India, with fifteen conditioning factors.
• Assessed factor appropriateness using multi-collinearity tests and LASSO, and evaluated model performance using accuracy, precision, F1-score, and AUC-ROC.

Main Results:

• The hybrid XGBoost model (HXGBRS) demonstrated the highest prediction accuracy (AUC=0.937, Precision=0.946, F1-score=0.926, Accuracy=89.92%).
• The integration of Rough Set theory notably improved the prediction capabilities of all evaluated hybrid models.
• The HXGBRS model exhibited superior stability and effectively avoided overfitting, outperforming other models like HBPNNRS, HBNRS, HBRS, and HRFRS.

Conclusions:

• The proposed HXGBRS hybrid model, enhanced with Rough Set theory, is highly effective for landslide prediction in the study area.
• The developed integrated GIS platform offers a dynamic and user-friendly tool for predicting landslide probability in large prone regions.
• This approach provides a valuable framework for assessing landslide impacts on slopes and monitoring national routes.