<![CDATA[The growth of sugarcane requires optimal environmental conditions and the availability of balanced nutrients. However, fulfilling nutrition is a challenge because it requires targeted observation. The study proposes a machine learning-based decision support model using a predictive empirical approach to monitor nutrient needs and recommend fertilizer dosages. The proposed approach integrates field data with a two-layer modeling framework to support fertilization decision-making. The classification model predicts the status of nutrient adequacy, while the regression model estimates the level of fertilizer application. The target label (y) is generated through feature extraction using a rule-based empirical formula derived from the threshold of agronomic parameters. The nutrients analyzed included macronutrients (nitrogen, phosphorus, potassium) and micronutrients (iron, zinc, copper). Model development involves selecting the best-performing algorithm using recall for classification and RMSE and R² for regression. The results of the cross-validation showed that the Gradient Boosting algorithm achieved the most consistent performance, with a recall of 0.99 during training and >0.98 in holdout testing. The regression model also showed low RMSE and high R² values, especially for micronutrient estimation. The proposed model contributes to data-driven fertilization optimization.]]>
Copyrights © 2026
