An integrated approach of hybrid ensemble machinelearning-based efficient seismic slope fragilityassessment and GIS mapping

Seismic slope failures are difficult to predict due to the probabilistic nature of soil properties and seismic loads. Additionally, high-fidelity simulation-based fragility assessment methods for soil slopes require substantial computational resources, making it time-consuming to link these results to geographic information systems (GIS). Thus, this study proposes a computationally efficient approach for seismic slope fragility assessment, maintaining high accuracy while reducing computational cost. Based on slope failure thresholds from observation data, extensive displacement-based fragility analyses are conducted, and High-Confidence-of-Low-Probability-of-Failure (HCLPF) values are calculated across diverse slope conditions. Based on such an HCLPF dataset, a machine learning (ML) model predicting HCLPF of fragility analyses is established. A Hybrid Ensemble Method (HEM) combining Extreme Gradient Boosting (XGB) and Bagging Ensemble Method (BEM) is newly proposed for accurate prediction. The sub-strategy is also presented to reduce iterative optimization cost for XGB within the proposed HEM. Consequently, the HEM model outperformed existing individual and ensemble models in accuracy on test data. Also, when integrated into GIS, the HEM-based fragility prediction map closely matched a high-fidelity simulation-based fragility map, achieving about 95% accuracy while reducing computational costs by about 96%.