همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

Interpretable Machine Learning for Rocking-Induced Settlement Prediction Using SHAP Analysis

نویسندگان :

Seyed Emad Miri¹ Hamid Mohammadnezhad²

1- دانشگاه شهید بهشتی 2- دانشگاه شهید بهشتی

کلمات کلیدی :

Machine Learning،Rocking Shallow Foundations،SHAP،Interpretability،Supervised Learning

چکیده :

Rocking shallow foundations have been increasingly recognized as an effective mechanism for seismic energy dissipation; however, their permanent settlement remains difficult to predict due to the complex soil-foundation interaction. This paper presents the development and evaluation of two supervised machine learning models, namely Extra Trees and LightGBM, using a dataset of 140 experimental cases. Based on learning curve analysis, the most optimized predictive behavior was achieved when 80% of the data was utilized during model training. Meanwhile, Extra Trees provided the best predictive performance on the test set, with a coefficient of determination of 0.945. To ensure interpretability rather than relying on black box behavior, the contribution of each input variable to the model output is investigated using Shapley Additive Explanations, describing how changes in each variable affect the predicted settlement. Accordingly, the slenderness ratio appears to be the most influential factor on rocking-induced settlement, followed by the rocking coefficient, maximum ground acceleration, and Arias intensity. On the other hand, the foundation area ratio and soil type have a relatively minor effect. The combination of ensemble learning models and Shapley Additive Explanations provides both strong predictive capability and transparent insights, therefore supporting more reliable and interpretable seismic design practices.