Enhanced slope stability prediction using ensemble machine learning techniques.

Area of Science:

• Geotechnical Engineering
• Machine Learning
• Computational Science

Background:

• Accurate slope stability prediction is crucial in geotechnical engineering but remains challenging.
• Existing methods often lack the precision and speed required for real-world applications.
• Machine learning offers potential for enhanced slope stability evaluation.

Purpose of the Study:

• To develop an ensemble machine learning model for accurate slope stability prediction.
• To evaluate the model's performance from both classification and regression perspectives.
• To compare the ensemble model's efficacy against base classifiers.

Main Methods:

• An ensemble bagging and boosting technique was employed with Decision Tree (DT) and Random Forest (RF) as base classifiers.
• Seven quantitative parameters were selected from 125 data points, utilizing random cross-validation.
• Dimensionality reduction techniques were applied to assess data aggregation impacts.

Main Results:

• Ensemble models achieved >90% accuracy in classification, outperforming base classifiers by 8-10%.
• Ensemble bagging regression improved R-squared values by 8-10% compared to conventional models.
• Random Forest and ensemble bagging with DT demonstrated superior robustness and accuracy, even after dimension reduction.

Conclusions:

• Ensemble machine learning models, particularly ensemble bagging with DT, provide a highly effective method for slope stability prediction.
• The proposed models offer a novel and accurate approach for geotechnical slope engineering.
• The ensemble bagging model is identified as the most effective for evaluating and predicting slope stability based on experimental results.