Enhanced Earth Slope Stability Assessment Using Computational Intelligence Algorithms

Proper slope stability prediction, particularly in mountainous terrains, is essential to reduce the catastrophic consequences of failures. Calculating the factor of safety (FS) with traditional methods is challenging and requires either tedious computations or sophisticated software. Advancements in machine learning (ML) methods and data collection have substantially improved slope stability analysis. While many ML algorithms have been applied to evaluate slope failures, there is a lack of a comprehensive comparative analysis across these algorithms with a broad dataset. This study employed classical and ML procedures to classify and predict the FS required for geotechnical design. Novel models were developed using a large dataset that included different soil and geometric attributes. The FS was modeled in terms of soil unit weight, cohesion, friction angle, slope angle and height, and pore water pressure ratio, using Python. Eight statistical metrics and confusion matrix measures were employed to evaluate the reliability of the models. The results showed that ML models were effective in predicting and classifying the FS, with the random forest model demonstrating optimal performance in terms of accuracy for both regression and classification models. The model's applicability was further confirmed with an independent validation dataset. Sensitivity analysis results indicated that soil cohesion was the most influential parameter on the FS, while slope height had the least impact. The illustrative example demonstrated the direct implementation of the model compared to traditional solutions. The findings of this study assist practitioners in estimating the FS required for the preliminary assessment of slope failure and in selecting appropriate protective measures and mitigation techniques. This can lead to better decision-making, optimized design processes, and increased sustainability for geotechnical and highway projects involving earth slopes.