Scholarlink

<p>Flood disasters in Kerala, India, have intensified due to changing precipitation patterns and rapid urbanization, with the 2018 floods causing 474 deaths and $3.8&#xa0;billion in damages. Current flood susceptibility mapping approaches often fail to address extreme class imbalance in spatial data and lack integration with socioeconomic vulnerability factors. The analysis processed 181.6&#xa0;million spatial data points across 18 flood conditioning factors, addressing severe class imbalance (511:1 ratio) through a novel two-stage resampling approach combining RandomUnderSampler and SMOTEENN. A Boruta-inspired feature selection reduced variables to 14 critical factors. Five machine learning algorithms, viz., Random Forest, XGBoost, LightGBM, CatBoost, and Histogram-based Gradient Boosting, were implemented and evaluated for their effectiveness in predicting flood susceptibility across the study region. All models achieved strong predictive capability, while XGBoost achieved the highest accuracy at 0.984, followed by LightGBM and CatBoost at 0.981, Histogram-Based Gradient Boosting at 0.980, and Random Forest at 0.968. Each model demonstrated robust predictive discrimination power through their respective ROC AUC scores that exceeded 0.98. The models maintained minimal false positive ratings between 0.027 and 0.054 and almost zero false negative results between 0.002 and 0.005. Classification methods facilitated susceptibility maps, integrated with socioeconomic factors through a four-component vulnerability framework (hazard, exposure, sensitivity, and adaptive capacity). The proposed analysis identified Malappuram (0.831) and Palakkad (0.719) are the most vulnerable districts. This research establishes novel methodologies for flood risk assessment applicable to disaster management and urban planning in regions with similar geographical characteristics in Kerala.</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

Improving Flood Susceptibility Assessment in Kerala: Integration of Advanced Machine Learning Models with Socioeconomic Vulnerability Analysis

  • 期刊论文

  • 发表日期：

    2026年3月4日

查看全文
  • M. S. Kendagannaswamy,
  • C. K. Roopa,
  • B. S. Harish,
  • M. S. Maheshan

摘要

Flood disasters in Kerala, India, have intensified due to changing precipitation patterns and rapid urbanization, with the 2018 floods causing 474 deaths and $3.8 billion in damages. Current flood susceptibility mapping approaches often fail to address extreme class imbalance in spatial data and lack integration with socioeconomic vulnerability factors. The analysis processed 181.6 million spatial data points across 18 flood conditioning factors, addressing severe class imbalance (511:1 ratio) through a novel two-stage resampling approach combining RandomUnderSampler and SMOTEENN. A Boruta-inspired feature selection reduced variables to 14 critical factors. Five machine learning algorithms, viz., Random Forest, XGBoost, LightGBM, CatBoost, and Histogram-based Gradient Boosting, were implemented and evaluated for their effectiveness in predicting flood susceptibility across the study region. All models achieved strong predictive capability, while XGBoost achieved the highest accuracy at 0.984, followed by LightGBM and CatBoost at 0.981, Histogram-Based Gradient Boosting at 0.980, and Random Forest at 0.968. Each model demonstrated robust predictive discrimination power through their respective ROC AUC scores that exceeded 0.98. The models maintained minimal false positive ratings between 0.027 and 0.054 and almost zero false negative results between 0.002 and 0.005. Classification methods facilitated susceptibility maps, integrated with socioeconomic factors through a four-component vulnerability framework (hazard, exposure, sensitivity, and adaptive capacity). The proposed analysis identified Malappuram (0.831) and Palakkad (0.719) are the most vulnerable districts. This research establishes novel methodologies for flood risk assessment applicable to disaster management and urban planning in regions with similar geographical characteristics in Kerala.