Scholarlink

<p>Rainfall prediction plays a vital role in climate-sensitive sectors such as agriculture, water resource management, and disaster preparedness. This study compares the performance of Holt-Winters exponential smoothing, a classical statistical model, with a deep learning-based Recurrent Neural Network using Long Short-Term Memory (RNN-LSTM) for monthly rainfall forecasting. Using historical rainfall data from Belagavi, India (1901–2002), both models were developed and evaluated on their predictive accuracy. The RNN-LSTM model was implemented using TensorFlow and Keras, with appropriate data preprocessing and hyperparameter tuning. Results revealed that the LSTM model significantly outperformed the Holt-Winters method, achieving lower mean absolute error (MAE: 3.23 vs. 6.31), lower mean absolute percentage error (MAPE: 1.04 vs. 0.89), and higher coefficient of determination (R<sup>2</sup>: 0.825 vs. 0.50) during the testing phase. This comparative analysis highlights the capability of deep learning models in capturing complex temporal dependencies in climatic data, providing a more reliable alternative to traditional forecasting approaches for Indian rainfall prediction.</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

Rainfall Prediction Comparison Between Holt Winter’s and Long Short-term Memory, A Deep Learning Technique

  • 期刊论文

  • 发表日期：

    2025年6月5日

查看全文
  • Shravankumar Shivappa Masalvad,
  • Sibghatullah Inyatullah Khan,
  • Anamika Yadav

摘要

Rainfall prediction plays a vital role in climate-sensitive sectors such as agriculture, water resource management, and disaster preparedness. This study compares the performance of Holt-Winters exponential smoothing, a classical statistical model, with a deep learning-based Recurrent Neural Network using Long Short-Term Memory (RNN-LSTM) for monthly rainfall forecasting. Using historical rainfall data from Belagavi, India (1901–2002), both models were developed and evaluated on their predictive accuracy. The RNN-LSTM model was implemented using TensorFlow and Keras, with appropriate data preprocessing and hyperparameter tuning. Results revealed that the LSTM model significantly outperformed the Holt-Winters method, achieving lower mean absolute error (MAE: 3.23 vs. 6.31), lower mean absolute percentage error (MAPE: 1.04 vs. 0.89), and higher coefficient of determination (R2: 0.825 vs. 0.50) during the testing phase. This comparative analysis highlights the capability of deep learning models in capturing complex temporal dependencies in climatic data, providing a more reliable alternative to traditional forecasting approaches for Indian rainfall prediction.