Prediction of slope stability using Tree Augmented Naive-Bayes classifier: modeling and performance evaluation

Abstract

Predicting slope stability is critical for identifying terrain that is prone to landslides and mitigating the damage caused by landslides. The relationships between factors that determine slope instability are complicated and multi-factorial, so it is sometimes difficult to mathematically characterize slope stability. In this paper, new Tree Augmented Naive-Bayes (TAN) model was developed to predict slope stability subjected to circular failures based on six input factors: cohesion, internal friction angle, pore pressure ratio, slope angle, unit weight, and slope angle. A total 87 slope stability case records obtained from published literature was used to train and test the proposed TAN model. According to the results of the performance indices—accuracy, precision, recall, F-score and Matthews correlation coefficient, the established TAN model was proven to be better at predicting slope stability with acceptable accuracy than other formerly developed empirical models in the literature. Furthermore, the slope height was revealed as the most sensitive factor in a sensitivity analysis.

Keywords

• machine learning algorithm
• tree augmented naive-bayes
• sensitivity analysis
• slope stability prediction