Development of an optimized deep learning model for predicting slope stability in nano silica stabilized soils

Area of Science:

• Geotechnical Engineering
• Artificial Intelligence
• Material Science

Background:

• Assessing infinite slope stability is crucial in geotechnical engineering.
• Traditional methods (LEM, FEM) are computationally intensive and struggle with non-linear soil stabilization effects.
• Nano-silica (NS) stabilization enhances soil properties and mechanical strength, necessitating advanced analysis techniques.

Purpose of the Study:

• To develop and validate a hybrid deep learning model for predicting the stability of Nano-silica (NS) stabilized infinite slopes.
• To optimize the model using the Optuna algorithm for enhanced predictive performance.
• To improve model interpretability using Explainable AI (XAI) and SHAP techniques.

Main Methods:

• A hybrid deep learning model integrating Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Recurrent Neural Networks (RNN) was developed.
• The model was optimized using the Optuna algorithm.
• Explainable AI (XAI) and SHAP techniques were applied for feature importance analysis.

Main Results:

• The optimized RNN-CNN-LSTM model achieved a 99.4% accuracy on unseen test data.
• Stable validation trends and robust predictive performance were observed.
• Cohesion (c), Nano-Silica Content (NS%), and Slope Angle (β) were identified as the most influential factors for slope stability.

Conclusions:

• Hybrid deep learning models, optimized and interpretable, offer a powerful and efficient tool for geotechnical engineers to assess slope stability.
• The proposed framework reduces computational effort and improves predictive accuracy compared to conventional methods.
• The model can be integrated into real-time early warning systems for enhanced landslide risk assessment and infrastructure resilience.