Soil liquefaction, a phenomenon triggered by seismic activity, poses significant risks to structures built on saturated, loose, cohesionless, or low-cohesion soils. Accurate prediction of liquefaction-induced settlements remains a critical challenge in geotechnical engineering due to the complex interplay of soil properties, site conditions, and seismic parameters. Traditional methods, including empirical and numerical approaches, are often limited in handling the inherent uncertainties and complexities of the problem. This study utilized a real-world dataset, leveraging explainable artificial intelligence (XAI) to predict liquefaction-induced settlements. Ensemble learning algorithms were employed to train the model, and the best-performing algorithm was further analyzed using the Local Interpretable Model-Agnostic Explanations (LIME) framework. The results demonstrate that thicker liquefiable layers, higher earthquake magnitudes, and greater PGA values significantly increase settlement, whereas deeper soil layers and higher standard penetration resistance reduce settlement. LIME provides transparent and interpretable insights into the contribution of various factors influencing settlement predictions, offering a novel perspective on the problem. The results demonstrate the potential of XAI to enhance the interpretability and reliability of predictive models, paving the way for more robust and practical solutions for geotechnical applications.
Dağdeviren et al. (Mon,) studied this question.