Prediction of slope failure probability based on machine learning with genetic-ADASYN algorithm

Abstract

In geotechnical engineering, precise probabilistic assessment of slope stability is crucial for risk management and the safe operation of engineering projects. To perform probabilistic assessments of slopes accurately and efficiently, six machine learning (ML) algorithms, including Logistic Regression (LR), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Extremely Randomized Tree (ERT), and Gradient Boosting Tree (GBT), are adopted to establish surrogate models for the relationship between a slope's safety factor and soil's shear parameters. Latin Hypercube Sampling (LHS) is employed to generate training samples for constructing surrogate models using ML algorithms. Adaptive Synthetic Sampling (ADASYN) is used to balance the number of samples of failure and safety classes by generating synthetic samples for the failure sample set, and a Genetic algorithm (GA) is used to optimize the hyper-parameters of ML and ADASYN algorithms to improve the accuracy of the surrogate models. Two criteria are proposed to measure the accuracy of surrogate models built using ML with Genetic-ADASYN algorithms, and a formula is presented to estimate the optimal number of samples for the training set. Based on the surrogate models, reliability indices and failure probabilities of slopes can be readily estimated using the Monte Carlo Simulation Method (MCSM). Case studies of five slopes with different complexities are adopted to illustrate the proposed method for the probabilistic analysis of slope stability and to compare the accuracy of surrogate models built using different ML algorithms. Results show that the Genetic-ADASYN algorithm can significantly improve the accuracy of surrogate models built using ML algorithms, and among the six ML algorithms, GBT is the best ML algorithm due to its generalizability and accuracy in slope stability prediction problems. The research findings can provide a reference for risk assessment of slope engineering and help to improve the accuracy and efficiency of probabilistic slope stability analysis.