Search for a command to run...
Fertilizers need to be utilized in the most efficient way possible to ensure that the farming process is more productive and less damaging to the environment. Traditional systems of fertilizer recommendations often focus on human experience and fixed recommendations, which cannot consider changing environmental and soil situations. The proposed system concerning the fertilizer recommendations applied with the use of machine learning is much more efficient in the reduction of fresh fertilizer application and thus more environmentally friendly. Based on a database of 9,999 farming records, comprising eight key parameters: temperature, humidity, moisture, crop type, soil type, nitrogen, potassium, and phosphorus, the presented system selects the most efficient among the seven possible types of fertilizers. The three models constructed and tested by the researchers were the Light Gradient Boosting Machine (LightGBM), the Extreme Gradient Boosting (XGBoost) and an ensemble of Decision Tree, Random Forest and LightGBM classifiers. The study of feature importance has shown that soil nutrients and climatic conditions were the most important factors in predicting fertilizers with an accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 4. 2 \%}$</tex>. LightGBM has been found to be more effective than XGBoost (89.4) and the ensemble model (84.9) with the greater accuracy. This is because the results indicate that application of boosting strategies may significantly improve making of decisions regarding the use of fertilizers using sustainable management. The developed system, by offering a scalable, intelligent, and easy-to-use framework, which can be incorporated with the real agricultural systems, allows the farmers the power to apply the fertilizers accurately.