Scholarlink

<p>The increasing intensity and influence of landslides through climate change require the application of sophisticated landslide susceptibility assessment methodologies with both geological and social vulnerability factors. Traditional models rely purely on geological data, failing to address social vulnerabilities that may be most critical in determining impact scenarios of disaster events. This study introduces a sophisticated predictive model through the inclusion of machine learning with analysis for social vulnerability for the purpose of a comprehensive risk assessment framework. This model utilizes the hybrid approach through the combination of Random Forest and Gradient Boosting for prediction accuracy enhancement, and a quad Long Short-Term Memory (LSTM) architecture to analyze time-series data based on critical factors like rainfall and soil moisture. Achieving high prediction accuracy at 92% accuracy, a precision of 0.89, recall at 0.91, and the F1-score as 0.90, while a quad LSTM architecture provides a temporal risk assessment with 85% accuracy. Application of PCA in conjunction with cluster analysis has been used in reducing the dimensionality of the data so that 90% of data variance is captured in five clusters and key susceptibility patterns are identified with a purity score of 0.88. This multivariate approach considerably enhances landslide susceptibility assessment, allowing focused disaster management and mitigation strategies.</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

Integrating social vulnerability to improve landslide susceptibility assessment quality: a hybrid machine learning approach

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月30日

查看全文
  • Prashant Pande,
  • Jayant Raut,
  • Rajesh Bhagat,
  • Sagar Dhengare

摘要

The increasing intensity and influence of landslides through climate change require the application of sophisticated landslide susceptibility assessment methodologies with both geological and social vulnerability factors. Traditional models rely purely on geological data, failing to address social vulnerabilities that may be most critical in determining impact scenarios of disaster events. This study introduces a sophisticated predictive model through the inclusion of machine learning with analysis for social vulnerability for the purpose of a comprehensive risk assessment framework. This model utilizes the hybrid approach through the combination of Random Forest and Gradient Boosting for prediction accuracy enhancement, and a quad Long Short-Term Memory (LSTM) architecture to analyze time-series data based on critical factors like rainfall and soil moisture. Achieving high prediction accuracy at 92% accuracy, a precision of 0.89, recall at 0.91, and the F1-score as 0.90, while a quad LSTM architecture provides a temporal risk assessment with 85% accuracy. Application of PCA in conjunction with cluster analysis has been used in reducing the dimensionality of the data so that 90% of data variance is captured in five clusters and key susceptibility patterns are identified with a purity score of 0.88. This multivariate approach considerably enhances landslide susceptibility assessment, allowing focused disaster management and mitigation strategies.