Scholarlink

<p>Early detection of cotton production variability offers a chance to make early decision, maximize yield and productivity, and boost profits for the farmers. The present research aimed to predict the seed cotton yield by using 536 germplasm accessions of <i>G. hirsutum</i>; through the data generated for consecutive two years. Machine learning tools (bootstrap forest, boosted tree, XGBoost and LightGBM), deep learning (artificial neural network) and multiple linear regression (MLR) models were used to forecast the seed cotton yield. The data set was divided between training and validation sets in a specified (75:25) ratio. A multiple linear regression model was developed to predict yield with prediction accuracy of 51% (R<sup>2</sup> = 0.51), using yield-attributing parameters as independent variables. The overall model comparison showed that the XGBoost model achieved the highest accuracy in terms of seed cotton yield prediction (R<sup>2</sup> = 0.93) with the lowest Root Average Square Error (RASE) and Absolute Average Error (AAE) followed by the LightGBM (R<sup>2</sup> = 0.79) and Bootstrap Forest model (R<sup>2</sup> = 0.78). The multiple linear regression showed the lowest accuracy (R<sup>2</sup> = 0.51) in comparison with other models. The XGBoost model demonstrated robust performance in categorical yield prediction also, achieving a precision of 0.66, a recall of 0.73 and an F1-score of 0.69, which validated the effectiveness of integrating phenotypic traits within a gradient boosting framework to accurately classify yield potential. Variable importance analysis showed irrespective of the models, number of bolls per plant, boll weight and field emergence were the major predictors of seed cotton yield. Optimization of the yield attributing parameters using the suitable models for predicting seed cotton yield will facilitate the future breeding programme developing the high yielding varieties and hybrids.</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

Application of machine learning tools for prediction of seed cotton yield in upland cotton (G. hirsutum L.) under irrigated ecosystem

  • 期刊论文

  • 发表日期：

    2026年6月24日

查看全文
  • Debashis Paul,
  • Vinita Gotmare,
  • M. Saravanan,
  • Sandeep Bagri,
  • Harish Kumbhalkar,
  • Subhash Chandra,
  • S. K. Verma,
  • Rishi Kumar,
  • Y. G. Prasad

摘要

Early detection of cotton production variability offers a chance to make early decision, maximize yield and productivity, and boost profits for the farmers. The present research aimed to predict the seed cotton yield by using 536 germplasm accessions of G. hirsutum; through the data generated for consecutive two years. Machine learning tools (bootstrap forest, boosted tree, XGBoost and LightGBM), deep learning (artificial neural network) and multiple linear regression (MLR) models were used to forecast the seed cotton yield. The data set was divided between training and validation sets in a specified (75:25) ratio. A multiple linear regression model was developed to predict yield with prediction accuracy of 51% (R2 = 0.51), using yield-attributing parameters as independent variables. The overall model comparison showed that the XGBoost model achieved the highest accuracy in terms of seed cotton yield prediction (R2 = 0.93) with the lowest Root Average Square Error (RASE) and Absolute Average Error (AAE) followed by the LightGBM (R2 = 0.79) and Bootstrap Forest model (R2 = 0.78). The multiple linear regression showed the lowest accuracy (R2 = 0.51) in comparison with other models. The XGBoost model demonstrated robust performance in categorical yield prediction also, achieving a precision of 0.66, a recall of 0.73 and an F1-score of 0.69, which validated the effectiveness of integrating phenotypic traits within a gradient boosting framework to accurately classify yield potential. Variable importance analysis showed irrespective of the models, number of bolls per plant, boll weight and field emergence were the major predictors of seed cotton yield. Optimization of the yield attributing parameters using the suitable models for predicting seed cotton yield will facilitate the future breeding programme developing the high yielding varieties and hybrids.