Interpretability of XGBoost in Imbalanced Seismic Performance Classification of Reinforced Concrete Buildings Using Resampling Techniques - 九大コレクション | 九州大学附属図書館

＜会議発表論文＞
Interpretability of XGBoost in Imbalanced Seismic Performance Classification of Reinforced Concrete Buildings Using Resampling Techniques

作成者	作成者名 Ryan James S. Olivo 所属機関所属機関名 Center for Structural Engineering and Informatics, MSU – Iligan Institute of Technology 所属機関所属機関名 Research Institute for Engineering and Innovative Technology, MSU – Iligan Institute of Technology 所属機関所属機関名 Department of Civil Engineering, University of Southern Mindanao
作成者	作成者名 Joel G. Opon 所属機関所属機関名 Center for Structural Engineering and Informatics, MSU – Iligan Institute of Technology 所属機関所属機関名 Research Institute for Engineering and Innovative Technology, MSU – Iligan Institute of Technology 所属機関所属機関名 Department of Civil Engineering and Technology, MSU – Iligan Institute of Technology
本文言語	英語
出版者	International Exchange and Innovation Conference on Engineering & Sciences
発行日	2025-10-30
収録物名	Proceedings of International Exchange and Innovation Conference on Engineering & Sciences (IEICES)
巻	11
開始ページ	443
終了ページ	449
会議情報	会議名 International Exchange and Innovation Conference on Engineering & Science(IEICES) 回次 11 主催機関 Interdisciplinary Graduate School Of Engineering Sciences (IGSES) Kyushu University 開催期間 2025-10-30～2025-10-31 開催会場 Kyushu University 九州大学開催地 Fukuoka 福岡開催国日本
出版タイプ	Version of Record
アクセス権	open access
Crossref DOI	https://doi.org/10.5109/7395554
権利関係	Copyright © 2025 IEICES/Kyushu University. All rights reserved.
権利関係	Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
概要	XGBoost is widely used in performance-based earthquake engineering (PBEE) because of its excellent performance, scalability, efficiency and ability to capture complex patterns. However, it often lack...s interpretability, which makes it harder to trust, especially when classifying the seismic performance of reinforced concrete buildings into Immediate Occupancy (IO), Life Safety (LS), and Collapse Prevention (CP). This issue becomes more difficult when the CP class is underrepresented. This study examines how three resampling methods: ADASYN (oversampling), EditedNN (undersampling), and SMOTE-Tomek (hybrid) affect how well we can understand XGBoost predictions. SHAP and Partial Dependence Plots (PDPs) are used to explain which features the model focuses on. Results show that oversampling reveals more influential and important features beyond Sa(1.0), such as We, SR, Eh, and Sa(T1), improving model transparency. Careful resampling, therefore, is critical. It is essential for interpretable and trustworthy models in PBEE.続きを見る

本文ファイル

ファイル	ファイルタイプ	利用条件	サイズ	閲覧回数	説明
2025_p0443	pdf	なし	3.87 MB	543

詳細

EISSN	2434-1436
レコードID	7395554
主題	Imbalance Classification
	Interpretability
	Seismic Performance Levels
	XGBoost
登録日	2025.12.15
更新日	2025.12.22