Learning a complex metabolomic dataset using random forests and support vector machines