Confidence intervals only make sense when you can calculate them with a precise formula, which is not the case here. The prediction error is an estimate. An error of 18% means that your result is rather precise. At 50% it were rather unprecise, at 100% it's worthless.