<![CDATA[Monitoring soil nutrient levels such as nitrogen (N), phosphorus (P), and potassium (K) is essential to support fertilizer efficiency and sustainable agricultural land management. However, commonly used laboratory-based analytical methods are time-consuming and costly. Therefore, alternative approaches that are more practical and efficient are needed. This study aimed to develop an Artificial Neural Network (ANN)-based system for predicting soil nutrient levels using soil physical parameters, namely pH, temperature, moisture content, and electrical resistance, as input variables. Data were collected from red-yellow podzolic soil subjected to different fertilization treatments. After normalization, the data were trained using an ANN model with four input nodes, two hidden layers (each consisting of five nodes), and one output node, employing the backpropagation algorithm and evaluating 27 combinations of activation functions. The training results showed coefficients of determination (R²) of 0.9642 for nitrogen, 1.0000 for phosphorus, and 0.9996 for potassium, with RMSE values of 0.0107, 10.5386, and 0.016457 and RRMSE values of 8.5048%, 0.79786%, and 1.581111%, respectively. During validation, R² values of 0.7218 (nitrogen), 0.6479 (phosphorus), and 0.6137 (potassium) were obtained. Nitrogen prediction exhibited good accuracy (RMSE 0.0222; RRMSE 15.54%), potassium prediction showed moderate accuracy (RMSE 0.2963; RRMSE 28.46%), while phosphorus prediction resulted in relatively high errors (RMSE 1066.77; RRMSE 80.98%), indicating the need for further model development.]]>
Copyrights © 2025
