The aim of this research is to implement multilayer perceptron (MLP) and radial basis function (RBF) artificial neural networks (ANNs) for the estimation of surface and mass radon exhalation rates (Es and Em), and effective radium content (CRa) values in soils. Three distinct MLPANN (MLPANN1, MLPANN2 and MLPANN3) and RBFANN (RBFANN1, RBFANN2 and RBFANN3) models were created for this. purpose: MLPANN1 and RBFANN1 models for predicting Es values, MLPANN2 and RBFANN2 models for predicting Em values, and MLPANN3 and RBFANN3 models for predicting CRa values in soil. The datasets for the training and testing phases of the generation and evaluation of the MLPANN and RBFANN models were generated using data obtained from the literature, including soil radon concentration (CRn), Es, Em, and CRa values. The verification and assessment of the MLPANN and RBFANN models were performed using the mean absolute error (MAE), root mean square error (RMSE), relative absolute error (RAE), ratio of RMSE to standard deviation of observations (RSR), and variance account for (VAF) statistics. After developing and checking all the MLPANN and RBFANN models, the best MLPANN and RBFANN models were selected. Each MLPANN and RBFANN model achieved over 81% accuracy (Pearson′s correlation coefficient = 0.81) in the training and testing phases, showing a very strong correlation between actual and predicted soil Es, Em, and CRa values. The validation results of the models showed that (a) the MLPANN1 and RBFANN1 models had similar and low MAE, RMSE, RAE, and RSR values and similar high VAF values, indicating that the MLPANN1 and RBFANN1 models predicted soil Es values with similar accuracy; and (b) MLPANN2 and MLPANN3 models had lower MAE, RMSE, RAE, and RSR values and higher VAF values than the RBAFANN2 and RBFANN3 models respectively, indicating that the MLPANN2 and MLPANN3 models outperformed the RBAFANN2 and RBFANN3 models in predicting soil Em and CRa values. The rank analysis of the MLPANN and RBFANN models revealed that (a) the MLPANN1 and RBFANN1 models had the same rank value in total, indicating their equally strong performance in predicting soil Es values; and (b) the MLPANN2 and MLPANN3 models had higher overall scores than the RBFANN2 and RBFANN3 models, respectively, indicating that the MLPANN2 and MLPANN3 models performed better in terms of prediction accuracy than the RBFANN2 and RBFANN3 models in predicting soil Em and CRa values, respectively. Taylor graphs showed that (a) both MLPANN1 and RBFANN1 models predicted soil Es values fairly accurately; and (b) MLPANN2 and MLPANN3 models predicted soil Em and CRa values more accurately than RBFANN2 and RBFANN3 models, respectively.

Optimization of artificial neural network for predicting radon exhalation rates and effective radium content in soil samples

期刊论文
发表日期：

2025年1月14日

查看全文

Selin Erzin

摘要

The aim of this research is to implement multilayer perceptron (MLP) and radial basis function (RBF) artificial neural networks (ANNs) for the estimation of surface and mass radon exhalation rates (E_s and E_m), and effective radium content (C_Ra) values in soils. Three distinct MLPANN (MLPANN1, MLPANN2 and MLPANN3) and RBFANN (RBFANN1, RBFANN2 and RBFANN3) models were created for this. purpose: MLPANN1 and RBFANN1 models for predicting E_s values, MLPANN2 and RBFANN2 models for predicting E_m values, and MLPANN3 and RBFANN3 models for predicting C_Ra values in soil. The datasets for the training and testing phases of the generation and evaluation of the MLPANN and RBFANN models were generated using data obtained from the literature, including soil radon concentration (C_Rn), E_s, E_m, and C_Ra values. The verification and assessment of the MLPANN and RBFANN models were performed using the mean absolute error (MAE), root mean square error (RMSE), relative absolute error (RAE), ratio of RMSE to standard deviation of observations (RSR), and variance account for (VAF) statistics. After developing and checking all the MLPANN and RBFANN models, the best MLPANN and RBFANN models were selected. Each MLPANN and RBFANN model achieved over 81% accuracy (Pearson′s correlation coefficient = 0.81) in the training and testing phases, showing a very strong correlation between actual and predicted soil E_s, E_m, and C_Ra values. The validation results of the models showed that (a) the MLPANN1 and RBFANN1 models had similar and low MAE, RMSE, RAE, and RSR values and similar high VAF values, indicating that the MLPANN1 and RBFANN1 models predicted soil E_s values with similar accuracy; and (b) MLPANN2 and MLPANN3 models had lower MAE, RMSE, RAE, and RSR values and higher VAF values than the RBAFANN2 and RBFANN3 models respectively, indicating that the MLPANN2 and MLPANN3 models outperformed the RBAFANN2 and RBFANN3 models in predicting soil E_m and C_Ra values. The rank analysis of the MLPANN and RBFANN models revealed that (a) the MLPANN1 and RBFANN1 models had the same rank value in total, indicating their equally strong performance in predicting soil Es values; and (b) the MLPANN2 and MLPANN3 models had higher overall scores than the RBFANN2 and RBFANN3 models, respectively, indicating that the MLPANN2 and MLPANN3 models performed better in terms of prediction accuracy than the RBFANN2 and RBFANN3 models in predicting soil E_m and C_Ra values, respectively. Taylor graphs showed that (a) both MLPANN1 and RBFANN1 models predicted soil E_s values fairly accurately; and (b) MLPANN2 and MLPANN3 models predicted soil E_m and C_Ra values more accurately than RBFANN2 and RBFANN3 models, respectively.