PREDICTION OF SUBSOILING DRAFT REQUIREMENTS OF SILT LOAM SOILS FROM EMPIRICAL DATA: ARTIFICIAL NEURAL NETWORK VERSUS MULTIPLE LINEAR REGRESSION

Abstract

Modern agriculture is intensive resulting in increased passes of machinery on soils leading to soil compaction. Wheel traffic makes the soil harder resulting in higher tillage draft requirement and fuel consumption. There is little documentation on prediction of subsoiling draft requirements for silt loam soils of Njoro Sub County, Nakuru County in Kenya. This study aimed to evaluate the performance of multiple linear regression (MLR) and Artificial Neural Networks (ANN) in predicting subsoiling draft requirement. Experimental data were collected from plots with wheel traffic treatments of 1, 2, 3, 4 and 5 passes to determine the effect of traffic intensity at three depths of 0 – 20 cm, 20 - 30 cm and 20 - 40 cm. The subsoiling draft was determined for each experiment and modeled using MLR and ANN. The results showed that ANN gave a more accurate prediction of the draft requirement of a subsoiler than an MLR. The RMSE for the models were 0.313 and 0.03 kN for the MLR and ANN model respectively. The MAPE for the MLR model was 5.41% while for ANN was 1.27%. The coefficient of determination for the two models were 0.984 and 0/999 respectively; ANN was therefore found to be suitable for predicting the subsoiling draft.