Scholarlink

Purpose <p>Selenium(Se) is an essential micronutrient for health and rice is the main source of Se for human body. However, rice grown in Se rich soil may not necessarily be Se rich, and rice grown in non Se rich soil may also be Se rich. Therefore, the purpose of this paper is to develop a machine learning algorithm to predict Se in rice and provide a basis for the division of Se-rich land.</p> Methods <p>30095 topsoil samples and 927 pairs of soil-rice samples were collected in Fujian province, China. The random forest(RF) and back-propagation artificial neural network(BP-ANN) model were selected and optimized. Based on the optimal hyperparameters, the two models were trained respectively, and the better one was selected by the reverse verification of the training results.</p> Results <p>The optimized results of the BP-ANN indicated that the RMSE value was the lowest when there were 6 neurons in the first hidden layer and 6 neurons in the second hidden layer, and the RMSE value was the lowest when the ntree and mtry values were 30 and 2 respectively in RF model. Based on the optimal hyperparameter, the R of RF(<i>R</i> = 0.93) was higher than BP-ANN(<i>R</i> = 0.81).</p> Conclusion <p>The RF had the better performance than BP-ANN. SOM, P, pH and MAP were the main factors affecting Se content in rice through feature importance analysis.</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

Prediction of soil-rice selenium transfer by machine learning

  • 期刊论文

  • 发表日期：

    2025年2月13日

查看全文
  • Ying Wang,
  • Junliang Cai,
  • Zhongfang Yang

摘要

Purpose

Selenium(Se) is an essential micronutrient for health and rice is the main source of Se for human body. However, rice grown in Se rich soil may not necessarily be Se rich, and rice grown in non Se rich soil may also be Se rich. Therefore, the purpose of this paper is to develop a machine learning algorithm to predict Se in rice and provide a basis for the division of Se-rich land.

Methods

30095 topsoil samples and 927 pairs of soil-rice samples were collected in Fujian province, China. The random forest(RF) and back-propagation artificial neural network(BP-ANN) model were selected and optimized. Based on the optimal hyperparameters, the two models were trained respectively, and the better one was selected by the reverse verification of the training results.

Results

The optimized results of the BP-ANN indicated that the RMSE value was the lowest when there were 6 neurons in the first hidden layer and 6 neurons in the second hidden layer, and the RMSE value was the lowest when the ntree and mtry values were 30 and 2 respectively in RF model. Based on the optimal hyperparameter, the R of RF(R = 0.93) was higher than BP-ANN(R = 0.81).

Conclusion

The RF had the better performance than BP-ANN. SOM, P, pH and MAP were the main factors affecting Se content in rice through feature importance analysis.