Interpretable deep learning models for predicting the bearing capacity of strip footings on cohesionless sloping soil

Purpose Accurately predicting the bearing capacity of shallow foundations on sloping soil is a critical challenge in geotechnical engineering due to the complex interactions between soil properties and structural parameters. This study employs deep learning models Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), and Residual Networks (ResNet) to enhance predictive accuracy and interpretability. Design/methodology/approach A dataset comprising 792 finite element simulations from Plaxis 2D was used to train and validate these models. Performance evaluation was conducted using Mean Absolute Error (MAE), Mean Squared Error (MSE), and R2 metrics. Among the models, CNN demonstrated the highest single-run accuracy, whereas ResNet exhibited the greatest stability across training iterations. To improve model transparency, SHapley Additive exPlanations (SHAP) were employed to analyze feature importance. Findings Results revealed that the footing depth-to-width ratio (Df/B) had the most significant impact on bearing capacity predictions, followed by slope inclination (ß) and internal friction angle (φ). ResNet captured more intricate feature interdependencies, while CNN and DNN relied on more direct relationships. The findings validate deep learning as a reliable approach for geotechnical predictions and highlight the importance of integrating explainable AI techniques to enhance trust in data-driven decision-making. Originality/value Despite the demonstrated potential of machine learning techniques in providing accurate and interpretable predictions for soil–structure interaction phenomena, significant gaps remain in optimizing and applying convolutional neural networks CNNs or Residual Network (ResNet) for predicting soil capacity. The present study addresses this shortfall by developing deep learning models integrated with state-of-the-art explainability tools. Building on the existing body of research, this study focuses on two main research gaps. First, it introduces advanced DNN, CNN, and ResNet architectures tailored to predict the bearing capacity of strip footings on cohesionless sloping soils. Second, it integrates SHAP to enhance the model interpretability, allowing for transparent and explainable predictions. The models were trained using a dataset derived from Plaxis analysis, considering key geotechnical parameters. By coupling deep learning architectures with interpretability tools, this research aims to enhance both model accuracy and transparency, thereby improving the practical applicability of AI-driven predictive tools in foundation design scenarios.