Using a relatively small amount of accessible data, we developed machine learning models to predict alfalfa yield and compared how different sets of features affected their error. We also compared the regression tree (RT), random forest (RF), neural network, support vector machine (SVR), k-nearest neighbors (KNN), Bayesian ridge regression, and linear regression methods. These methods were trained and evaluated with cross validation. The best set of features consisted of the Julian day of the harvest, the number of days between the sown date and the harvest date, and the cumulative amount of solar radiation and rainfall the crop received since the previous harvest. The RF, KNN, RT, and SVR methods obtained results that, when averaged, did not vary significantly from each other. The best individual model was a RF with a R2 of 0.941. This model had the highest R2 value compared to the best results from similar studies.
