Performance Prediction Evaluation of Machine Learning Models for Slope Stability Analysis: A Comparison Between ANN, ANN-ICA and ANFIS

Abstract

Slope stability analysis is crucial in civil engineering for the design and maintenance of embankments, especially those constructed on soft soils. Traditional methods like the limit equilibrium method (LEM) and finite element method (FEM) are time-consuming and require significant expertise. This study explores the application of three machine learning models—Artificial Neural Network (ANN), ANN combined with Imperialist Competitive Algorithm (ANN-ICA), and Adaptive Neuro-Fuzzy Inference System (ANFIS)—to predict slope stability. A numerical analysis using PLAXIS 2D software generated a database encompassing various geometric characteristics such as slope height, surcharge, and slope angle. These features served as input parameters, while the factor of safety (FOS) values were used as target outputs. The performance of each model was evaluated using determination coefficients (R²) and root mean square errors (RMSE). The ANN-ICA hybrid model demonstrated superior predictive accuracy, with R² and RMSE values of 0.998 and 0.041 for training datasets, respectively, outperforming the standalone ANN (R² = 0.724, RMSE = 0.124) and ANFIS (R² = 0.858, RMSE = 0.052) models. This study highlights the potential of hybrid machine learning approaches in enhancing the efficiency and accuracy of slope stability predictions, offering a promising alternative to traditional methods.