Scholarlink

<p>This study presents an interpretable, rule-based framework for analyzing structural damage patterns associated with the 2023 Kahramanmaraş earthquake sequence by integrating supervised and unsupervised machine learning approaches. A binary station-neighborhood–based damage label is employed to represent spatial damage patterns, enabling regional-scale assessment using site-specific and earthquake-related parameters derived from open-source datasets. In the supervised framework, decision tree models evaluated through nested leave-one-out cross-validation achieved an accuracy of 86.36% and an F1-score of 88.46%, indicating robust generalization performance given the limited dataset. The model trained on the full dataset yielded higher training accuracy (95.45%) and F1-score (96.00%), reflecting its ability to capture interpretable decision rules. The results consistently identify engineering bedrock depth (EBD) as the most informative predictor within the selected parameter space, with the horizontal-to-vertical spectral ratio (A<sub>0</sub>) providing additional discriminatory capability in resolving classification ambiguities. Complementary insights obtained from association rule mining (confidence = 1.0, support = 0.25) further highlight the importance of EBD, Vs30, and predominant frequency (f₀), while the selected strong-motion parameters, peak ground acceleration (PGA), epicentral distance (Repi), and rupture distance (Rrup), contributed comparatively less to predictive performance within the adopted feature space. These findings indicate that, under conditions where widespread structural deficiencies provide a baseline level of vulnerability, site-condition proxies exhibited stronger statistical association with the observed station-neighborhood damage labels within the adopted feature space. The proposed framework demonstrates the potential of combining interpretable machine learning with site-condition indicators to support regional-scale damage prioritization and risk-informed decision-making, particularly in data-limited environments. Furthermore, the approach provides a foundation that can be extended by incorporating advanced ground-motion parameters to enhance post-earthquake damage assessment and disaster management applications.</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

Rule-based damage assessment of the 2023 Kahramanmaraş earthquakes: a decision tree and association analysis approach

  • 期刊论文

  • 发表日期：

    2026年7月21日

查看全文
  • Mustafa Senkaya,
  • Mustafa E. Okursoy,
  • Tülin İnkaya,
  • Ali Silahtar,
  • Hasan Karaaslan

摘要

This study presents an interpretable, rule-based framework for analyzing structural damage patterns associated with the 2023 Kahramanmaraş earthquake sequence by integrating supervised and unsupervised machine learning approaches. A binary station-neighborhood–based damage label is employed to represent spatial damage patterns, enabling regional-scale assessment using site-specific and earthquake-related parameters derived from open-source datasets. In the supervised framework, decision tree models evaluated through nested leave-one-out cross-validation achieved an accuracy of 86.36% and an F1-score of 88.46%, indicating robust generalization performance given the limited dataset. The model trained on the full dataset yielded higher training accuracy (95.45%) and F1-score (96.00%), reflecting its ability to capture interpretable decision rules. The results consistently identify engineering bedrock depth (EBD) as the most informative predictor within the selected parameter space, with the horizontal-to-vertical spectral ratio (A0) providing additional discriminatory capability in resolving classification ambiguities. Complementary insights obtained from association rule mining (confidence = 1.0, support = 0.25) further highlight the importance of EBD, Vs30, and predominant frequency (f₀), while the selected strong-motion parameters, peak ground acceleration (PGA), epicentral distance (Repi), and rupture distance (Rrup), contributed comparatively less to predictive performance within the adopted feature space. These findings indicate that, under conditions where widespread structural deficiencies provide a baseline level of vulnerability, site-condition proxies exhibited stronger statistical association with the observed station-neighborhood damage labels within the adopted feature space. The proposed framework demonstrates the potential of combining interpretable machine learning with site-condition indicators to support regional-scale damage prioritization and risk-informed decision-making, particularly in data-limited environments. Furthermore, the approach provides a foundation that can be extended by incorporating advanced ground-motion parameters to enhance post-earthquake damage assessment and disaster management applications.