Computational Model to Predict Soil Density Using Machine Learning Methods

The machine learning methods have been used successfully in the calculation of parameters of various problems of engineering, in which the complicated variables have a relation nonlinear among themselves and the modelation does not enable representing the intervening problem through a mathematical function of easy deduction. For the estimation of soil properties several variables are analyzed that make their estimation by means of mathematical models is a complex process transferring the problem solution to artificial intelligence field. The present work aims at developing a mathematical model for the estimation of soil density through the on-the-go soil sensing, a method of automatized learning. The computational learning automated tool used was WEKA, by means of which three procedures of automatized learning applied (multilayer perceptron neuronal artificial nets and K-nearest neighbor). The validation of the model came true by means of the crossed and experimental validation. Results evidence that the best method is the K-nearest neighbor with absolute mean error of 0.06 and a correlation coefficient of 0.89; variables of bigger weight in prediction were moisture content followed by work speed, power, width of the working tool and the depth.