<![CDATA[A study comparing the performance of machine learning algorithms to predict soil nutrient values based on soil impedance has been conducted. The algorithm models used include Linear Model, K-Nearest Neighbors (K-NN) with n-neighbors 3, 18, 21, 24, 27, and 30, Decision Tree with max depth 3, and Random Forest with n-estimators 6 and 21. During the training phase, 10 model variations with the best performance were found, including Linear Model, K-NN (n-neighbors), Decision Tree (max depth 3), and Random Forest (n-estimators 6 and 21). In the testing phase, Random Forest (n-estimator 21) showed the best performance with MAE = 0.15%, MSE = 0.09%, RMSE = 0.31%, and accuracy = 99.85%. Regression analysis indicated an R-squared value of 0.924, indicating that most of the variations in soil impedance values can be explained by variations in soil nutrient values. A regression value approaching 1 indicates that the regression model used has a very good ability to explain the variations observed in the data. This indicates that most of the variations in the dependent variable (the variable being predicted, which is the nutrient values) can be explained by the independent variable (the predictor variable, which is the soil impedance values) in the model. Correlation analysis resulted in a strong negative correlation between impedance and Al, Fe, K, Ca, Zn, Ni, Ta, V, Cr, and Mn (values -0.81 to -0.99), while a positive correlation occurred with Mg, Si, S, Cl, Ti, Zr, and Ga (values 0.65 to 0.99). This indicates that an increase in impedance values is generally followed by an increase in nutrient values.]]>
Copyrights © 2024
