The potential of machine learning methods for the prediction of soil compaction

The practice of precision farming is contingent upon a comprehensive understanding of the spa-tial variability of a multitude of physical and chemical soil parameters. The acquisition of knowledge regarding soil parameters necessitates the undertaking of soil sampling and subsequent analysis, a process that is inherently labour-intensive and time-consuming. Consequently, preci-sion farming employs the identification of homogeneous field regions through the utilisation of scanning techniques, with the objective of ascertaining soil electrical characteristics, including electrical conductivity and magnetic susceptibility. The objective of this study was to attempt to predict soil compaction based on selected electrical parameters. In order to predict compaction, machine learning methods, namely decision tree and support vector regression were employed. The highest R-value of 0.87 was obtained for the decision tree model and soil layer 0.1-0.2 m for the training set. For the test set, the highest R-value of 0.85 was obtained for soil layer 0.1-0.2 m and the support vector regression model, which also had the lowest MAPE error value of 11.31%. The prediction of soil compaction using electrical soil parameters based on machine learning methods represents a promising avenue of research.