An Intelligent Fertilizer Dosing System Using a Random Forest Model for Precision Agriculture

Asto Bonifacio R.F. · Journal of Robotics and Control (JRC) (2025) · DOI: 10.18196/jrc.v6i3.26421

Resumen

The inefficient application of fertilizers in horticultural crops, particularly in rural areas of Peru, leads to significant economic losses, soil degradation, and environmental risks. In response to this issue, this paper proposes an intelligent fertilizer dosing system that integrates solid and liquid fertilization applications through a predictive machine learning model. The main contribution of this research is the development and partial validation of an embedded system that dynamically adapts nutrient (NPK) doses based on real-time soil conditions, crop type, and phenological stage. The predictive model, based on Random Forest (RF), was trained using 10000 synthetic data points generated via Sobol-LHS sampling and validated with 1000 real field measurements. The method incorporates thirteen agronomic variables, including soil moisture, pH, temperature, and nutrient content, enabling adaptive control of the dosing mechanisms. The system achieved promising results, with root mean square errors (RMSE) of 2.81 kg/ha for nitrogen, 1.42 kg/ha for phosphorus, and 0.94 kg/ha for potassium. These results demonstrate the model’s ability to deliver accurate crop-specific fertilization recommendations, reducing input waste and improving nutrient use efficiency. Although full field trials are planned for future phases, the proposed system offers a scalable and low-cost solution for precision agriculture in resource-constrained settings, promoting more sustainable farming practices and enhancing the productivity of smallholder farmers.

Instituciones: Universidad Continental