Vol 19 no.1 2019
Department of Mathematical Sciences, Olabisi Onabanjo University, Ago-Iwoye, Nigeria1,4, Department of Physical and Computer Sciences, McPherson University, Seriki Sotayo, Nigeria2, Department of Information and Communication Technology, Osun State University, Osogbo, Nigeri3
Intensifying sustainable agriculture and management of natural resources can be achieved with digital mapping of soil functional properties in data sparse regions of Africa. This research is aimed at evaluating different multi-output Regression models (Random Forest Regressor (RFR), Linear Regressor (LR), Extremely Randomized Trees Regressor (ET) and Bagging Regressor using LR as base classifier (BLR))to predict five soil properties (Calcium (Ca), Phosphorus (P), Potential of Hydrogen (pH), Soil Organic Carbon (SOC) and Sand of different soil sample) simultaneously on African Soil sample dataset. Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Coefficient of Correlation (R2) and Explained Variance metrics were used to evaluate the performances of these models on the dataset. The result obtained revealed that RFR performed best based on RMSE, MAE and Explained Variance that other models. LR performed inferior to other models but its ensembles with BLR improve its predictive performance.