Predicting Africa Soil Properties Using Machine Learning Techniques

Different machine learning algorithms were assessed for estimating five functional soil parameters (SOC content, Calcium content, Phosphorous content, sand content, and pH value). The algorithms used include variants of linear regression and support vector regression. A closer look at the prediction performance for each target revealed that apart from pH, which consistently had worse performance, prediction for the other soil properties was quite satisfactory (RMSE < 0.4). Applying machine learning techniques to soil properties prediction has shown a lot of promising and encouraging results. Getting more data, domain knowledge and intuition, possibly from soil scientist/experts, would surely maximize this potential for accurate soil property prediction.