Fertilizer Recommender using XGBoost and Random Forest

Abstract

Agricultural productivity in Zambia and many developing countries is significantly affected by poor soil fertility management and the lack of site-specific fertilizer recommendations. Traditional fertilizer advisory methods often provide generalized guidance, resulting in nutrient imbalance, reduced crop yield, increased production costs, and environmental degradation. This paper presents a Fertilizer Recommender System using a hybrid XGBoost and Random Forest (XG-RF) machine learning approach to generate personalized fertilizer recommendations based on soil nutrients, crop type, weather conditions, and historical farm data. The system was developed using the Kaggle Fertilizer Recommendation Dataset containing soil parameters such as nitrogen (N), phosphorus (P), potassium (K), pH, moisture, temperature, rainfall, crop type, fertilizer type, and dosage information. Random Forest was selected for its robustness and ability to handle non-linear agricultural data, while XGBoost was employed for its high predictive accuracy and efficient gradient boosting capability. The dataset was preprocessed, normalized, and divided into training and testing sets using an 80:20 ratio, with model evaluation performed using accuracy, precision, recall, and F1-score metrics. Experimental results demonstrated that the proposed XG-RF model outperformed traditional standalone approaches, achieving improved fertilizer prediction accuracy and optimized fertilizer dosage recommendations. The system also incorporates continuous learning through farmer feedback and historical records to improve future predictions. By integrating precision agriculture techniques with machine learning, the proposed solution supports sustainable farming practices, reduces fertilizer misuse, minimizes environmental impact, and enhances crop productivity for major Zambian crops including maize, soybean, cassava, tomato, and groundnut.