Scholarlink

<p>Archaeological Predictive Model is crucial for efficient site identification but faces challenges like high uncertainty in negative samples, limited accuracy, and simplistic analyses of site-environment interactions. This study, based on 189 Kushan-period sites in the Surkhandarya Region of Uzbekistan, introduces a kernel density estimation (KDE)-based strategy to improve negative sample selection. Five machine learning models, incorporating geomorphological, climatic, and terrain variables, were assessed for predicting site locations. SHAP (SHapley Additive exPlanations) analysis was used to investigate the relationship between environmental factors and settlement patterns. Results show that the proposed strategy significantly enhances predictive performance, with AUC and accuracy increasing by 12.1% and 14%, respectively. Random Forest outperformed other models in robustness across various conditions. Land cover, slope, and precipitation emerged as key factors influencing site distribution. This research offers a novel framework for archaeological prospection, combining machine learning with environmental analysis.</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

Explainable artificial intelligence with negative sample optimization for archaeological site prediction in Surkhandarya Uzbekistan

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月31日

查看全文
  • Jia Yang,
  • Lei Luo,
  • Jianghong Zhao,
  • Dechang Ji,
  • Jisi Sun,
  • Jinhui Fan,
  • Xingjian Fu,
  • Ran Tu,
  • Xinyuan Wang

摘要

Archaeological Predictive Model is crucial for efficient site identification but faces challenges like high uncertainty in negative samples, limited accuracy, and simplistic analyses of site-environment interactions. This study, based on 189 Kushan-period sites in the Surkhandarya Region of Uzbekistan, introduces a kernel density estimation (KDE)-based strategy to improve negative sample selection. Five machine learning models, incorporating geomorphological, climatic, and terrain variables, were assessed for predicting site locations. SHAP (SHapley Additive exPlanations) analysis was used to investigate the relationship between environmental factors and settlement patterns. Results show that the proposed strategy significantly enhances predictive performance, with AUC and accuracy increasing by 12.1% and 14%, respectively. Random Forest outperformed other models in robustness across various conditions. Land cover, slope, and precipitation emerged as key factors influencing site distribution. This research offers a novel framework for archaeological prospection, combining machine learning with environmental analysis.