Boosting Algorithm Comparisons for the Prediction of Added Amount Macronutrients (NPK) of Harumanis Mango Tree Penology Stage

Abstract

The Harumanis mango, a prized cultivar grown in Perlis, Malaysia, requires meticulous nutrient management to enhance yield and fruit quality. Conventional soil nutrient analysis techniques are often expensive and time-consuming, highlighting the need for efficient predictive methods. This study explores the application of boosting algorithms to predict the added amount of NPK fertilizer macronutrient nitrogen (N), phosphorus (P), and potassium (K) critical for mango cultivation. The predictive models were developed based on soil nutrient data collected via TDR sensors throughout different Harumanis mango phenology stages. These data-driven models provide a cost-effective alternative to traditional soil testing, facilitating timely and precise nutrient management. To evaluate model performance, multiple boosting algorithms, including XGBoost, LightGBM, Gradient Boosting Regressor (GBR), and AdaBoost, were fine-tuned and assessed using performance metrics such as MAE, RMSE, R², RMSLE, and MAPE. Among these, the XGBoost model exhibited the highest predictive accuracy, achieving an MAE of 38.4046, RMSE of 51.6798, R² of 0.8278, RMSLE of 0.4507, and MAPE of 0.5739. The results indicate that the XGBoost model effectively forecasts soil nutrient levels, outperforming other evaluated models. Accurately predicting macronutrient concentrations enables targeted fertilization strategies, reducing costs and environmental impact while optimizing Harumanis mango production. However, the model relies on soil nutrient data and is highly dependent on accurate sensor readings. Future studies should focus on expanding the dataset and incorporating additional environmental parameters to further enhance model precision and applicability across diverse agricultural regions.