Machine learning in soil science for prediction and management of biological activity for sustainable land use

Abstract

The article discusses the use of machine learning methods for predicting and managing soil biological activity, which is a key aspect of sustainable land use. The development of a random forest model for predicting the Respiration parameter based on data on the physical and chemical characteristics of the soil collected in various areas of Baltimore, Maryland is shown. The model has demonstrated an accuracy of about 70%, which highlights its potential for application in the agricultural sector. The results of visualization of the distribution of actual and predicted values, as well as the analysis of prediction errors are presented. Prospects for further improvement of the model using a genetic algorithm to optimize hyperparameters and integrate additional data such as climatic conditions and historical land use data are discussed. The findings highlight the importance of using machine learning to improve agricultural production efficiency and minimize environmental impacts.