Predicting the MJO using interpretable machine-learning models