The procedure you followed does not work in this case, since the Ac recoils mostly fall below the ion transport threshold, and therefore they are not intercepted by USRYIELD.
Use RESNUCLEi instead, and divide the resulting Ac yields in the TARGET region by the aforementioned number of proton stars and multiply then by the reaction cross section reported in the USRYIELD output (2012 mb).
This way you shall get the actual Ac production cross section values of FLUKA that are as follows:
Lastly, the percentage you refer to is indeed the relative error estimation, but, when it’s so large, it indicates that the corresponding value and the error itself are poorly reliable.