Comparisons of Four Machine Learning Algorithms for Stability Evaluations of Highway Rock Slopes

Assessing the stability of slopes for a highway construction project is a challenging endeavor owing to the tight project schedule and complicated field geological conditions. This paper compared the performances of four machine learning algorithms, i.e. artificial neural network, support vector machine, decision tree, and random forest in predicting the instabilities of rock slopes along a highway in Anhui Province, China. Eight evaluation indicators were identified as the key parameters affecting the stability of the slopes at the study area, namely slope width, excavation height, average angle of excavated slope gradient, rock hardness characteristics, rock mass connectivity, stability of rock mass combined, rock mass integrity, and slope structure. The datasets of 80 slopes within the study area were extracted, filtered, transformed and dimensionally reduced. The data were divided into training set (64 samples) and test set (16 samples). The results showed that the prediction model based on the RF algorithm has outperformed the other three algorithms. The RF model has a better prediction accuracy and generalization ability, yielding a prediction accuracy as high as 0.938.