Scholarlink

<p>Floods pose serious threats to society and environmental sustainability and are among the most significant hazards worldwide. While Machine Learning (ML) models contain valuable tools to aid in flood prediction, they often operate as black boxes, making it difficult to interpret how they arrive at their decisions. In this study, we employ Genetic Programming (GP) to enhance the interpretability of ML-based flood models. The research focuses on Flood Susceptibility Mapping (FSM) using standalone GP and Logistic Regression (LR) algorithms, using GP integrated with the LR (GP-LR), and further with Principal Component Analysis (PCA) for dimensionality reduction (PCA-GP-LR). Data from 461 flooded and 461 non-flooded locations and thirteen spatial flood-influencing factors in Jahrom County, Iran, are utilized for spatial prediction. Flood susceptibility maps are developed and compared using four models: GP, LR, GP-LR, and PCA-GP-LR. Evaluation of these maps reveals that the GP-LR model achieved the highest accuracy, surpassing the GP, LR, and PCA-GP-LR models, based on the quality indicator Area Under the Receiver Operating Characteristic Curve (AUROC). We demonstrate that while combining GP with LR enhances performance compared to using the standalone GP model, the GP model's accuracy declines when combined with the PCA technique.</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

Advancing flood susceptibility mapping using evolutionary genetic programming and empirical models

  • 期刊论文

  • 发表日期：

    2026年2月23日

查看全文
  • Ali Pourzangbar,
  • Seyed Vahid Razavi-Termeh,
  • Abolghasem Sadeghi-Niaraki,
  • Mário J. Franca,
  • Soo-Mi Choi

摘要

Floods pose serious threats to society and environmental sustainability and are among the most significant hazards worldwide. While Machine Learning (ML) models contain valuable tools to aid in flood prediction, they often operate as black boxes, making it difficult to interpret how they arrive at their decisions. In this study, we employ Genetic Programming (GP) to enhance the interpretability of ML-based flood models. The research focuses on Flood Susceptibility Mapping (FSM) using standalone GP and Logistic Regression (LR) algorithms, using GP integrated with the LR (GP-LR), and further with Principal Component Analysis (PCA) for dimensionality reduction (PCA-GP-LR). Data from 461 flooded and 461 non-flooded locations and thirteen spatial flood-influencing factors in Jahrom County, Iran, are utilized for spatial prediction. Flood susceptibility maps are developed and compared using four models: GP, LR, GP-LR, and PCA-GP-LR. Evaluation of these maps reveals that the GP-LR model achieved the highest accuracy, surpassing the GP, LR, and PCA-GP-LR models, based on the quality indicator Area Under the Receiver Operating Characteristic Curve (AUROC). We demonstrate that while combining GP with LR enhances performance compared to using the standalone GP model, the GP model's accuracy declines when combined with the PCA technique.