Machine Learning-Based Slope Failure Prediction Model Considering the Uncertainty of Prediction

Abstract

Slope failure is a severe natural disaster that can cause property damage and human costs. In order to develop a warning system for slope failure, various studies have been conducted, including research based on both physics-based models and machine learning-based models. While machine learning-based approaches have shown promise due to their ability to automatically extract hidden patterns in data, conventional machine learning models have their limitations. Specifically, while they can always provide a prediction value, they fail to provide information about the uncertainty of the prediction results. In this study, we developed a machine learning model that can predict the slope failure by training trends in time-series data. Our proposed model addresses the limitations of the conventional machine learning models by incorporating the Monte Carlo dropout to calculate the uncertainty during the prediction stage. The experimental results demonstrated that the proposed model significantly outperforms the conventional machine learning models in terms of both its prediction accuracy and the ability to estimate uncertainty. Furthermore, the model proposed in this study can support decision-makers by providing more accurate information than the conventional models.

Keywords

• slope failure
• machine learning
• uncertainty