Prediction and analysis of nutrient values in soil by artificial intelligence methods: An example from isparta

The rapid increase in the world population has always constituted a significant problem, while natural resources have remained limited. Especially during periods such as the Industrial Revolution, the rate of increase has become even higher, making the responsibility to protect nature even more crucial. Preventing practices such as unconscious fertilization, excessive use of agricultural chemicals, and unnecessary irrigation are among the fundamental measures that can be taken. In this study, it was aimed to guide farmers towards conscious agricultural practices based on soil analysis. In this context, a portable Arduino-based NPK (Nitrogen, Phosphorus, Potassium) sensor device was developed. During the training phase, 5 803 real soil analysis data samples obtained from the Isparta Provincial Directorate of Special Administration were used. Using the pH, phosphorus, and potassium data collected from the soil via the sensor device, the values for saturation, lime, and organic matter were estimated with random forest, bagging, logistic regression, decision tree, and XGBoost algorithms. The XGBoost algorithm provided the highest accuracy rates for saturation, lime, and organic matter, with values of 77.07, 98.89, and 93.35, respectively. The obtained predictions were presented to farmers through a user-friendly web interface. Through this system, it is aimed to both increase agricultural productivity and support sustainable agriculture.