Scholarlink

<p>Early and accurate yield prediction of widely cultivated Maize (<i>Zea mays</i> L.) crop is essential for food security. The present study evaluates the maize cultivation in Katihar district, Bihar, India, during the Rabi season of year 2023–2024, focusing on crop acreage and yield prediction. This study incorporates indices derived from remote sensing images such as Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Green Chlorophyll Index (GCI), and Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) with weather factors (temperature, precipitation, and soil moisture) and historical yield data of twenty-two years (2002–2024). An unsupervised <i>k</i>-means clustering algorithm applied for image classification showed that the total cultivated area of maize crops is 35,293 hectares. In addition, four machine learning (ML) based regression models, optimized using random search, such as AdaBoost, Random Forest (RF), Automatic Relevance Determination (ARD), and Bayesian Ridge Regression (BRR) are applied to predict maize yield. With the lowest Mean Squared Error (MSE) of 0.08 t ha<sup>− 1</sup> and R<sup>2</sup> of 0.983, RF regression outperformed the others, forecasting an average maize crop yield of 9.46 t ha<sup>− 1</sup> for the 2023–2024 Rabi Season. The results demonstrate that integrating vegetation indices with weather data significantly enhances the accuracy of crop yield predictions. However, the results are based on limited seasonal observations due to the nature of agriculture-based data availability. This study aimed at evaluating the potential applicability of ensemble ML models rather than establishing definitive predictive performance. This approach provides valuable insights for stakeholders, researchers, and policymakers, supporting the adoption of data-driven strategies for effective crop management and food security planning.</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

Random Search Based Hyperparameter Optimization of Machine Learning Regression Models for Maize Crop Yield Prediction in an Agriculturally Prominent Landscape of India

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Priyanka Yadav,
  • Varun Narayan Mishra,
  • Maya Kumari,
  • Rajendra Prasad,
  • Sandeep Kumar Kaushik

摘要

Early and accurate yield prediction of widely cultivated Maize (Zea mays L.) crop is essential for food security. The present study evaluates the maize cultivation in Katihar district, Bihar, India, during the Rabi season of year 2023–2024, focusing on crop acreage and yield prediction. This study incorporates indices derived from remote sensing images such as Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Green Chlorophyll Index (GCI), and Fraction of Absorbed Photosynthetically Active Radiation (FAPAR) with weather factors (temperature, precipitation, and soil moisture) and historical yield data of twenty-two years (2002–2024). An unsupervised k-means clustering algorithm applied for image classification showed that the total cultivated area of maize crops is 35,293 hectares. In addition, four machine learning (ML) based regression models, optimized using random search, such as AdaBoost, Random Forest (RF), Automatic Relevance Determination (ARD), and Bayesian Ridge Regression (BRR) are applied to predict maize yield. With the lowest Mean Squared Error (MSE) of 0.08 t ha− 1 and R2 of 0.983, RF regression outperformed the others, forecasting an average maize crop yield of 9.46 t ha− 1 for the 2023–2024 Rabi Season. The results demonstrate that integrating vegetation indices with weather data significantly enhances the accuracy of crop yield predictions. However, the results are based on limited seasonal observations due to the nature of agriculture-based data availability. This study aimed at evaluating the potential applicability of ensemble ML models rather than establishing definitive predictive performance. This approach provides valuable insights for stakeholders, researchers, and policymakers, supporting the adoption of data-driven strategies for effective crop management and food security planning.