Scholarlink

<p>This study uses both statistical and machine learning methods to provide a comprehensive analysis of ragi (finger millet) cultivation trends in India. The Mann-Kendall test and Sen’s Slope estimator identify significant long-term trends in area, production, and yield, revealing a consistent decline in cultivation area (− 32.46 thousand ha/year) and production (− 10.00 thousand tonnes/year<b>)</b>, alongside a steady increase in yield (+ 16.30&#xa0;kg/ha/year<b>).</b> Unlike prior national-scale assessments, this study explicitly integrates trend diagnostics, machine learning prediction, and time-series forecasting into a reproducible hybrid framework. To forecast future yields, the study employs Random Forest and Gradient Boosting models, which achieve high predictive accuracy (R<sup>2</sup> = 0.91 and 0.94; RMSE = 94.3 and 77.5, respectively), and capture complex, non-linear relationships between input variables. These models are interpreted to reveal structural dependencies, moving beyond black-box prediction and enhancing transparency in agricultural modeling. By integrating trend diagnostics with ensemble learning, and comparing ARIMA and SARIMA forecasts to distinguish structural stability from seasonal variability, this work introduces a novel, scalable, and interpretable framework for district-level yield forecasting. The dual-model logic and interpretable ML integration offer methodological relevance beyond local contexts, contributing to international discourse on climate-resilient crop modeling. The findings highlight the urgency of enhancing productivity amidst shrinking cultivation areas and offer actionable insights for climate-resilient agricultural planning and policy formulation.</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

Predictive modeling of finger millet yield in India: an interpretable ensemble–time series framework

  • 期刊论文

  • 发表日期：

    2026年4月3日

查看全文
  • V. Guhan,
  • K. Nagaratna,
  • Dharma Raju Akasapu

摘要

This study uses both statistical and machine learning methods to provide a comprehensive analysis of ragi (finger millet) cultivation trends in India. The Mann-Kendall test and Sen’s Slope estimator identify significant long-term trends in area, production, and yield, revealing a consistent decline in cultivation area (− 32.46 thousand ha/year) and production (− 10.00 thousand tonnes/year), alongside a steady increase in yield (+ 16.30 kg/ha/year). Unlike prior national-scale assessments, this study explicitly integrates trend diagnostics, machine learning prediction, and time-series forecasting into a reproducible hybrid framework. To forecast future yields, the study employs Random Forest and Gradient Boosting models, which achieve high predictive accuracy (R2 = 0.91 and 0.94; RMSE = 94.3 and 77.5, respectively), and capture complex, non-linear relationships between input variables. These models are interpreted to reveal structural dependencies, moving beyond black-box prediction and enhancing transparency in agricultural modeling. By integrating trend diagnostics with ensemble learning, and comparing ARIMA and SARIMA forecasts to distinguish structural stability from seasonal variability, this work introduces a novel, scalable, and interpretable framework for district-level yield forecasting. The dual-model logic and interpretable ML integration offer methodological relevance beyond local contexts, contributing to international discourse on climate-resilient crop modeling. The findings highlight the urgency of enhancing productivity amidst shrinking cultivation areas and offer actionable insights for climate-resilient agricultural planning and policy formulation.