落石
可解释性
超参数
计算机科学
机器学习
人工智能
数据挖掘
地质学
地震学
山崩
作者
Haijia Wen,Jiwei Hu,Jialan Zhang,Xuekun Xiang,Mingyong Liao
标识
DOI:10.1061/9780784484982.011
摘要
Common machine learning has limited application in rockfall susceptibility mapping by a lack of interpretability. Represented by SHAP, explainable machine learning has been developed recently. This study proposed a novel interpretable hybrid-optimized model based on SHAP and XGBoost to interpret rockfall susceptibility evaluation results at both global and local levels. After hybrid-optimized by grid searching hyperparameters and recursive feature elimination (RFE) screening factors, only nine main factors selected by RFE from the 23 initial conditioning factors influence the occurrence of rockfall in the study area. The developed rockfall susceptibility evaluation model provided the accuracy, precision, and AUC value improved by 0.0846, 0.0809, and 0.0616, respectively, for the test data sets. The hybrid-optimized model has a good performance for rockfall susceptibility. The SHAP-based global and local interpretation could further insight into the rockfall occurrence mechanics.
科研通智能强力驱动
Strongly Powered by AbleSci AI