Slime Mould Algorithm Enhanced Gradient Boosting Regressor for Prediction of Slope Stability

Abstract

Present research details the comparison of predictive efficiency of linear regression, gradient boosting regression, and slime mould algorithm optimised gradient boosting regression. These algorithms have been used to predict the factor of safety of cut slopes in lower Tons valley, Uttarakhand, India. Initially, 103 soil slopes were examined for their stability factor in finite difference code. Four parameters like cohesion, angle of internal friction, slope height, and slope angle were selected to develop the stability (factor of safety) prediction models. Based on the statistical accuracy indices like R2 and mean absolute error, the hyper-parameter optimized model based on slime mould algorithm (i.e., SMA-GBR) had the best prediction capability. And, the R2 for SMA-GBR model is 0.92 and mean absolute error is 0.063. The next better performer is linear regression model with value of 0.84 and 0.094 for R2 and mean absolute error respectively. And, the gradient boosting algorithm model has the least predictive capacity as compared to other two prediction models developed in the present work. However, its capability is significantly increased as its hyper-parameter are fine-tuned through slime mould algorithms (SMA). © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.