Scholarlink

<p>Climate change in Bangladesh is causing unstable temperatures and rainfall, increasing the risk of floods, droughts, and agricultural disruption. Previous studies missed integrating temperature and rainfall data with feature selection and Explainable AI (XAI). This paper introduces a novel approach that integrates machine learning (ML) and deep learning (DL) methods with feature selection and XAI to improve climate impact predictions across the region. Using a comprehensive dataset from the Bangladesh Statistical Yearbook (1981–2023), our feature analysis highlights that key months such as April, February, and June strongly influence temperature and rainfall predictions, while location plays a greater role in shaping rainfall patterns than temperature. Our proposed ensemble model outperformed traditional and previous models across three climate datasets, achieving <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({\textrm{R}}^{2}\)</EquationSource> </InlineEquation> scores of 1.00, 0.79, and 0.99 for maximum temperature, minimum temperature, and rainfall, respectively. It also recorded low MAE values of 0.08, 0.39, and 3.46, and RMSE values of 0.10, 0.50, and 4.20 correspondingly. XAI techniques like LIME improved prediction interpretability by revealing key factors influencing the model’s decisions. Critical influences included lower March and July temperatures for maximum temperature predictions, while higher January and February temperatures positively affected the model. For rainfall, significant contributions from July and September rainfall were observed, while lower rainfall in April and December negatively impacted predictions. These insights support better agriculture and water management by identifying key climate drivers. Future work can enhance the model by adding more meteorological variables and considering large-scale climate patterns for deeper regional insights.</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

Predicting Climate Change Impacts on Temperature and Rainfall in Bangladesh Using Ensemble Learning: A Statistical Yearbook Analysis

  • 期刊论文

  • 发表日期：

    2026年6月18日

查看全文
  • Shahriar Siddique Ayon,
  • Md. Ebrahim Hossain,
  • Abdul Mubin Mahe,
  • Md Saef Ullah Miah,
  • M. Mostafizur Rahman

摘要

Climate change in Bangladesh is causing unstable temperatures and rainfall, increasing the risk of floods, droughts, and agricultural disruption. Previous studies missed integrating temperature and rainfall data with feature selection and Explainable AI (XAI). This paper introduces a novel approach that integrates machine learning (ML) and deep learning (DL) methods with feature selection and XAI to improve climate impact predictions across the region. Using a comprehensive dataset from the Bangladesh Statistical Yearbook (1981–2023), our feature analysis highlights that key months such as April, February, and June strongly influence temperature and rainfall predictions, while location plays a greater role in shaping rainfall patterns than temperature. Our proposed ensemble model outperformed traditional and previous models across three climate datasets, achieving \({\textrm{R}}^{2}\) scores of 1.00, 0.79, and 0.99 for maximum temperature, minimum temperature, and rainfall, respectively. It also recorded low MAE values of 0.08, 0.39, and 3.46, and RMSE values of 0.10, 0.50, and 4.20 correspondingly. XAI techniques like LIME improved prediction interpretability by revealing key factors influencing the model’s decisions. Critical influences included lower March and July temperatures for maximum temperature predictions, while higher January and February temperatures positively affected the model. For rainfall, significant contributions from July and September rainfall were observed, while lower rainfall in April and December negatively impacted predictions. These insights support better agriculture and water management by identifying key climate drivers. Future work can enhance the model by adding more meteorological variables and considering large-scale climate patterns for deeper regional insights.