Predictive modelling of sweep's specific draft using machine learning regression approaches

Abstract

Modelling and optimizing soil‐tool interaction parameters for tillage operations is crucial for developing efficient and precise tools. This study focused on a specific draft of sweep tools in the soil bin filled with vertisol, considering factors such as tool geometry, cone index, working depth and operational speed. Data analysis showed that the range of specific draft values, from 9.51 to 38.95 kN/m2. Machine learning models, including artificial neural network (ANN), support vector machine (SVM), bagged trees (BT) and boosted trees (BoT), were developed using experimental data to predict the specific draft of sweep tools with hyperparameter configuration. The developed machine learning models have also been compared with the predictive multiple linear regression (MLR) model, and it was found that the predictive performance of the machine learning models was better than the MLR model during training and testing. The fine‐tuned ANN model achieved impressive statistical performance with the lowest mean absolute error (MAE) of 0.489 kN/m2, root mean square error (RMSE) of 0.619 kN/m2, standard error of prediction (SEP) of 3.462% and highest coefficient of determination (R2) of .99 during testing. R2 values for BT, BoT, SVM and MLR models were .97, .96, .94 and .83, respectively, for specific draft predictions. The findings from this study have practical implications for optimizing sweep tool design and improving tillage operation. Manufacturers and farmers can benefit from predictive modelling using machine learning to design and select appropriate tillage tools for specific soil conditions. This approach can lead to improved soil health, increased yields and reduced costs.