Explainable AI using ensemble machine learning with integrated SHapley additive explanations (SHAP)-Borda approach for estimation of the safety factor against soil liquefaction

• Published in: Environmental earth sciences Vol. 84; no. 17; p. 507
• Main Authors: Dağdeviren, Uğur, Demir, Alparslan Serhat, Erden, Caner, Kökçam, Abdullah Hulusi, Kurnaz, Talas Fikret
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.09.2025; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Artificial intelligence; Biogeosciences; Decision making; Earth and Environmental Science; Earth Sciences; Earthquakes; Engineering; Environmental Science and Engineering; Explainable artificial intelligence; Geochemistry; Geology; Hydrology/Water Resources; Laboratories; Learning algorithms; Liquefaction; Machine learning; Methods; Original Article; Parameterization; Performance evaluation; Predictions; Safety factors; Soil; Terrestrial Pollution; Ensemble machine learning; SHapley Additive exPlanations (SHAP); Borda; Soil liquefaction; Prediction
• ISSN: 1866-6280; 1866-6299
• DOI: 10.1007/s12665-025-12466-z

In most of the studies on soil liquefaction prediction based on Machine Learning (ML), the models presented are presented in a closed box structure. In the studies where the effect of the features on the model performance is analyzed with Interpretability methods, it is seen that the order of effect of the features changes for each ML algorithm. This situation makes the results of the studies conducted on the same subject inconsistent. In this study, we propose an integrated SHapley Additive exPlanations (SHAP)-Borda approach to overcome this problem. With this study, we provide decision makers with ease in explaining ML models by combining SHAP analysis results with the Borda method for the first time. In the study, ensemble ML algorithms were used for soil liquefaction prediction using data collected from the literature. The performances of the model predictions obtained by hyper parameterization were compared, and correlation results ranging from 0.91 to 0.93 were obtained. Ensemble ML algorithms that were found to be successful as a result of evaluating other performance criteria were analyzed with the SHAP-Borda approach in the study. It has been observed that with the proposed SHAP-Borda approach, the interpretability results of different ML algorithms can be brought together, and a final result can be presented, providing ease of evaluation for decision makers. The study also shows that (N 1 ) 60 and a max are the most effective features in predicting soil liquefaction.