Estimation of leaf Mg values of apple trees in early period with machine learning methods

This study employs machine learning methods to estimate magnesium (Mg) levels in apple leaves at mid-vegetation from early period values. Leaf samples were collected from 150 apple orchards over two years, both at early and mid-vegetation stages. Soil samples were taken at 0–30 cm below the canopy. Leaf analyses included N, P, K, Ca, Mg, Fe, Cu, Mn, Zn, and B, while soil analyses covered pH, salinity, lime, organic matter, sand, clay, and silt percentages. The mid-vegetation Mg value of the leaves was the target variable, and other parameters served as predictors. The top five machine learning models (Nu support vector regressor, random forest regressor, histogram-based gradient boosting regression tree, K-nearest neighbors regressor, Bayesian ridge) were utilized. Using feature selection, it was found that reliable predictions could be made with five out of 17-input variables (Mg21, lime, Ca21, P21, sand). The models achieved R² values of approximately 0.59 with categorical data and 0.51 without, with corresponding MAE values of 0.03 and 0.04. These results, compared with existing literature, confirm that machine learning is an effective tool for early-stage Mg status prediction in apple trees.