Scholarlink

<p>Landslide susceptibility assessment (LSA) is crucial for effective regional geohazard risk management and territorial spatial planning. Recent advancements in machine learning (ML) have enhanced the accuracy of landslide susceptibility modeling. However, ML models face challenges in efficiently identifying potential landslide-prone areas because of the considerable time and resources required for data processing, model selection, and tuning, even for experienced specialists. This study proposes a hybrid approach that combines an automated ML model (AutoML) with the Shapley Additive exPlanations (SHAP) method to automatically predict landslides and provide a comprehensive explanation of the predictions. The AutoML model was employed for the southern foothills of Changbai Mountain, utilizing a dataset of 381 landslides with 16 landslide conditioning factors. The model automatically completed the assessment task, achieving an area under the curve (AUC) value of 0.9. Compared to traditional models, such as logistic regression, eXtreme Gradient Boosting, and two-dimensional convolutional neural network models, the AutoML model demonstrated high efficiency and good accuracy. Furthermore, the SHAP method offered a nuanced understanding of the interplay between model outcomes and conditioning factors from global and local perspectives. The SHAP method identified roads, altitudes, and slopes as key drivers of landslides in the case areas, revealing a non-linear relationship between feature variables and landslide prediction. These findings suggest that hybrid AutoML-SHAP can provide a new method for automatically assessing landslide susceptibility and introduce explainable artificial intelligence models.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Landslide susceptibility assessment using AutoML-SHAP method in the southern foothills of Changbai Mountain, China

  • 期刊论文

  • 发表日期：

    2025年2月18日

查看全文
  • Defeng Zheng,
  • Yuanyuan Li,
  • Chenglin Yan,
  • Hao Wu,
  • Yosuke Alexandre Yamashiki,
  • Botong Gao,
  • Tingkai Nian

摘要

Landslide susceptibility assessment (LSA) is crucial for effective regional geohazard risk management and territorial spatial planning. Recent advancements in machine learning (ML) have enhanced the accuracy of landslide susceptibility modeling. However, ML models face challenges in efficiently identifying potential landslide-prone areas because of the considerable time and resources required for data processing, model selection, and tuning, even for experienced specialists. This study proposes a hybrid approach that combines an automated ML model (AutoML) with the Shapley Additive exPlanations (SHAP) method to automatically predict landslides and provide a comprehensive explanation of the predictions. The AutoML model was employed for the southern foothills of Changbai Mountain, utilizing a dataset of 381 landslides with 16 landslide conditioning factors. The model automatically completed the assessment task, achieving an area under the curve (AUC) value of 0.9. Compared to traditional models, such as logistic regression, eXtreme Gradient Boosting, and two-dimensional convolutional neural network models, the AutoML model demonstrated high efficiency and good accuracy. Furthermore, the SHAP method offered a nuanced understanding of the interplay between model outcomes and conditioning factors from global and local perspectives. The SHAP method identified roads, altitudes, and slopes as key drivers of landslides in the case areas, revealing a non-linear relationship between feature variables and landslide prediction. These findings suggest that hybrid AutoML-SHAP can provide a new method for automatically assessing landslide susceptibility and introduce explainable artificial intelligence models.