Scholarlink

<p>The Western Ghats region of India experiences frequent landslides due to heavy downpours, rugged and unstable topography, population pressure, and unscientific tourism-related development initiatives. The Idukki district, the Western Ghats’ most severely battered district, with more than 1000 landslides in the 2018 monsoon season alone, was selected as the study area for landslide susceptibility modelling. This modelling aims to determine the effectiveness of various machine learning (ML) algorithms in identifying landslide susceptibility and to pinpoint the main landslide-conditioning factors. However, ML algorithms alone cannot tune their default hyperparameters—which control the models’ behavior—and they require support from optimization algorithms to improve predictability. Thus, this study integrated the particle swarm optimization (PSO) algorithm into four ML models: K-nearest neighbor (KNN), artificial neural network (ANN), support vector machine (SVM), and random forest (RF), to test whether the integration improved performance. The modelling employed 14 conditioning factors: seven topographic, three hydrological, and four environmental factors. The area under the curve (AUC) values for all eight models were over 0.80 (80%), indicating good performance. It was found that the RF model (AUC: 86.82%) had the highest performance, followed by ANN (AUC: 86.11%), SVM (AUC: 84.91%), and KNN (AUC: 82.84%) models. The integration of the PSO algorithm improved the performance of all four models, ranking them as follows: RF-PSO (AUC: 89.54%) &gt; SVM-PSO (AUC: 87.07%) &gt; ANN-PSO (AUC: 85.70%) &gt; KNN-PSO (AUC: 85.17%).Thus, it is ascertained that integrating the PSO algorithm enhanced the performance of all four models, with the RF-PSO model achieving the highest performance among the eight (four standalone and four optimized) models. The output will be indispensable for policymakers in enforcing bylaws and regulations related to future infrastructure development and land-use planning, thereby helping to minimize future impacts.</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 the spatial prediction of machine learning-based landslide susceptibility models by integrating the particle swarm optimization algorithm

  • 期刊论文

  • 发表日期：

    2025年9月15日

查看全文
  • Rajendran Shobha Ajin,
  • Anik Saha,
  • Sunil Saha,
  • Debi Prasanna Kanungo

摘要

The Western Ghats region of India experiences frequent landslides due to heavy downpours, rugged and unstable topography, population pressure, and unscientific tourism-related development initiatives. The Idukki district, the Western Ghats’ most severely battered district, with more than 1000 landslides in the 2018 monsoon season alone, was selected as the study area for landslide susceptibility modelling. This modelling aims to determine the effectiveness of various machine learning (ML) algorithms in identifying landslide susceptibility and to pinpoint the main landslide-conditioning factors. However, ML algorithms alone cannot tune their default hyperparameters—which control the models’ behavior—and they require support from optimization algorithms to improve predictability. Thus, this study integrated the particle swarm optimization (PSO) algorithm into four ML models: K-nearest neighbor (KNN), artificial neural network (ANN), support vector machine (SVM), and random forest (RF), to test whether the integration improved performance. The modelling employed 14 conditioning factors: seven topographic, three hydrological, and four environmental factors. The area under the curve (AUC) values for all eight models were over 0.80 (80%), indicating good performance. It was found that the RF model (AUC: 86.82%) had the highest performance, followed by ANN (AUC: 86.11%), SVM (AUC: 84.91%), and KNN (AUC: 82.84%) models. The integration of the PSO algorithm improved the performance of all four models, ranking them as follows: RF-PSO (AUC: 89.54%) > SVM-PSO (AUC: 87.07%) > ANN-PSO (AUC: 85.70%) > KNN-PSO (AUC: 85.17%).Thus, it is ascertained that integrating the PSO algorithm enhanced the performance of all four models, with the RF-PSO model achieving the highest performance among the eight (four standalone and four optimized) models. The output will be indispensable for policymakers in enforcing bylaws and regulations related to future infrastructure development and land-use planning, thereby helping to minimize future impacts.