‘부지화’ 잎의 SPAD측정값을 기반으로 잎의 질소함량을 추정하고자 여러 머신러닝 모델을 적용해 보았다. 모델 평가지표 및 실측치·예측치 데이터 산포도를 종합적으로 고려할 때, GB가 가장 적합한 모델로 선정되었다. 결정계수가 가장 1에 가까우며, MSE, RMSE, MAE도 모두 0에 수렴하여 실측치와 예측치의 오차가 가장 적었음을 알 수 있었다.

In recent years, to investigate the nitrogen content in plant leaves, the use of non-destructive and simple methods is preferred to that of destructive, time-consuming, and expensive methods. In this study, several machine learning models (linear and polynomial regressions, stochastic gradient descent, artificial neural network, support vector machine, k-nearest neighbors, random forest, and gradient boosting) were applied to estimate the nitrogen content in leaves based on the linear relationship between the SPAD reading value and the nitrogen content in leaves of Shiranuhi (Citrus unshiu × C. sinensis). As the data of nitrogen content measured under the laboratory condition was insufficient, the data was increased using the bootstrapping method. Considering the model evaluation metrics, the gradient boosting model was selected as the most accurate model. The coefficient of determination of this model was the closest to 1 and the MSE, RMSE, and MAE all converged to 0, indicating that the error between the measured and predicted values was the smallest.