Rock slope stability analysis using ensemble decision tree approaches and feature importance along an economic corridor in central India

Abstract

Large-scale slope destabilization poses significant risks, particularly during rapid infrastructure development along key economic corridors. The present study provides an advanced analysis of rock slope stability along a crucial route, National Expressway-4 connecting Mumbai and New Delhi, a region characterized by geologically complex terrain. Utilizing the Hoek-Brown criterion within a Finite Element Method (FEM) framework, the study simulates Strength Reduction Factors (SRF) under various conditions, emphasizing the influence of the Geological Strength Index (GSI). A comprehensive dataset varying seven critical input parameters was generated from these simulations. Machine learning (ML) algorithms, particularly tree-based models, were employed to predict SRF values. The Random Forest (RF) model emerged as the most accurate, achieving an $R²$ value of 0.9704, a root means square error of 0.2045, and a mean absolute error of 0.0526. Other models, like Gradient Boosting (GB) and eXtreme Gradient Boosting (XGBoost), also performed well but were slightly less accurate. The analysis highlighted the significant impact of slope height, angle, and GSI on model predictions by feature importance analysis and visualized through Radar plots. Later a rating system for important parameters was proposed based on research findings. This study demonstrates the effectiveness of integrating field data, FEM analysis, and machine learning techniques for assessing slope stability, with the Random Forest model proving particularly robust in identifying vulnerable slopes along this critical economic corridor.