# Automated slope stability assessment using modified Morgenstern-Price method and machine learning integration

## TL;DR

This paper introduces a hybrid system combining a geotechnical method with machine learning to quickly and accurately assess slope stability under various conditions.

## Contribution

The study develops an automated MP-LEM–ML framework for finite slope stability analysis with pore pressure and seismic effects.

## Key findings

- Slope angle, height, and cohesion were the main factors affecting the Factor of Safety.
- CatBoost achieved the highest predictive accuracy with R2 = 0.999 and RMSE = 0.014.
- The MP-LEM implementation was verified with published benchmarks, achieving R2 = 0.94.

## Abstract

This paper proposes a new hybrid automated framework that combines the Morgenstern-Price Limit Equilibrium Method (MP-LEM) with machine learning to predict the Factor of Safety (FSlope) for finite slopes under static and seismic loading conditions. The first synthetic data generation involved approximately 100,000 slope conditions that were varied through key geotechnical and seismic parameters: unit weight (γ), slope height (H), cohesion (c′), friction angle (ϕ′), slope angle (β), horizontal and vertical seismic coefficients (kh and kv), pore pressure ratio (µ), and the interslice force-scaling factor (λ). For model development, a stratified random sample of 20,000 cases was collated to maintain computational feasibility while preserving distributional characteristics. Corresponding FSlope values were computed using a simplified force-equilibrium form of the MP-LEM. The novelty of this study lies in the development of a comprehensive automated MP-LEM–ML framework specifically designed for finite slope stability analysis, which integrates a simplified Morgenstern–Price formulation with machine-learning models while explicitly accounting for pore pressure effects, bi-directional seismic loading, and interslice force scaling. The framework also automates the hyperparameter tuning, model evaluation, and optimal model selection within a reproducible Python environment, which enables a practical and deployable system for rapid finite slope stability assessment. Sensitivity analysis found that slope angle (β), slope height (H), and cohesion (c′) were the dominant controls on FSlope. Of nine tested ML algorithms (ANN, RF, DT, KNN, XT, GBoost, AdaBoost, XGBoost, and CatBoost), the CatBoost model had the highest predictive accuracy: R2 = 0.999, RMSE = 0.014 on the test set. The MP-LEM implementation was verified further by comparisons to published benchmark results. It returned R2 = 0.94, confirming its appropriateness as a reliable deterministic reference. Although the automated framework represents significant computational advantage, it is conditioned by the synthetic dataset based on the 2D and homogeneous soil assumptions, and by the underrepresentation of high-FSlope cases. Future work should focus on considering real case histories, probabilistic extensions, and 3D effects for further enhancement of practical applicability. In summary, the hybrid framework proposed herein represents a promising fast, accurate, and scalable decision-support framework for geotechnical slope stability assessment.

The online version contains supplementary material available at 10.1038/s41598-026-38670-w.