Enhancing rock slope stability prediction using random forest machine learning: A case study

ABSTRACT: The prediction of slope stability is a complex nonlinear problem. This paper proposes a new method based on the random forest (RF) algorithm to study the rocky slopes stability. Taking the Bukit Merah, Perak and Twin Peak (Kuala Lumpur) as the study area, the slope characteristics of geometrical parameters are obtained from a multidisciplinary approach (consisting of geological, geotechnical, and remote sensing analyses). 18 factors, including rock strength, rock quality designation (RQD), joint spacing, continuity, openness, roughness, filling, weathering, water seepage, temperature, vegetation index, water index, and orientation, are selected to construct model input variables while the factor of safety (FOS) functions as an output. The area under the curve (AUC) value of the receiver operating characteristic (ROC) curve is obtained with precision and accuracy and used to analyse the predictive model ability. With a large training set and predicted parameters, an area under the ROC curve (the AUC) of 0.95 is achieved. A precision score of 0.88 is obtained, indicating that the model has a low false positive rate and correctly identifies a substantial number of true positives. The findings emphasise the importance of using a variety of terrain characteristics and different approaches to characterise the rock slope.