Abstract: The complex nature of slope engineering presents challenges in accurately predicting slope stability with traditional methods. Identifying the appropriate techniques for slope stability prediction is essential in mitigating the risks associated with slope failures. This study conducts a thorough analysis of two boosting machine learning models: Adaboost and LightGBM. By evaluating a wide range of hyperparameters, the research aims to discover the optimal settings for each model, ultimately leading to effective solutions.Six potentially relevant features were identified as key indicators for prediction: height (H), pore water ratio (ru), unit weight (Ƴ), cohesion (c), slope angle (β), and angle of internal friction (ɸ). The models were assessed using evaluation indicators such as AUC and accuracy, revealing that LightGBM significantly outperformed the Adaboost model, achieving an impressive AUC of 0.878 and an accuracy of 0.803. Furthermore, real-world engineering examples illustrate the effectiveness of LightGBM as a predictive tool for slope stability. Its enhanced capacity and efficiency in deformation prediction positions it as a leading instrument for accurate forecasting in this field. To deepen the understanding of these models, a comprehensive analysis of parameter sensitivity was also conducted, highlighting the most significant characteristics contributing to reliable slope stability predictions.
Keywords: adaboost, lightgbm, slope stability, finite element, machine learning.
Title: Utilizing Boosting Machine Learning Techniques for Slope Stability Prediction
Author: Saurabh Kumar Anuragi
International Journal of Civil and Structural Engineering Research
ISSN 2348-7607 (Online)
Vol. 12, Issue 2, October 2024 - March 2025
Page No: 53-58
Research Publish Journals
Website: www.researchpublish.com
Published Date: 04-December-2024
