Abstract
Assessing the stability of infinite slopes is a critical challenge in geotechnical engineering, particularly by introducing Nano-silica (NS) stabilization, which changes soil properties and increases mechanical strength. Traditional methods of analysis of slope stability, such as the methods of limit equilibrium methods (LEM) and the methods of finite elements (FEM), often require extensive computational resources and effectively do not capture non-linear relations between soil stabilization and mechanisms of inclination failure. This study suggests a hybrid classification model of deep learning, integration of convolutional neural networks (CNN), long short-term memory (LSTM), and recurrent neural networks (RNN), optimized by Optuna to predict the stability of NS stabilized infinite slope. The model's training and verification were used with the data file containing 3,159 cases of inclination with different percentages of NS, and geotechnical parameters. The results show that RNN-CNN-LSTM, optimized through OPTUNA algorithms, overcomes conventional machine learning models and achieves an accuracy of 99.4% on unseen test data, supported by stable validation trends and robust predictive performance. Soil Index (SI), Unit Weight (γ), Curing Days (CD), Nano-Silica Content (NS%), Cohesion (c), Internal Friction Angle (Ø), Slope Height (H), Slope Angle (β), Pore Water Pressure Ratio (r(u)) were the features used for Explainable Artificial Intelligence (XAI) and SHAP (Shapley Additive Explanations). Furthermore, XAI and SHAP techniques were employed to enhance model interpretability, revealing that features c, NS%, and β are the most influential factors governing slope stability. This research shows that hybrid models of deep learning combined with techniques of optimization and interpretability provide a powerful and efficient tool for geotechnical engineers to assess the stability of inclination, reduce computing efforts, and improve predictive accuracy. The proposed framework can be integrated into early warning systems and monitoring platforms in real time, increasing the assessment of risks and infrastructure resistance in regions susceptible to landslides.