You could try a different database. RDP does not include environmental clusters, which can be a substantial proportion of amplicons. Perhaps see if you get more classification when using GreenGenes or SILVA
It sounds to me pretty reasonable, depending on where your samples are from (there is more unknown in prairie soil that mammal gut for instance). If your data are sequenced amplicons like I suspect, did you "denoise" and filter chimera as a preliminary step? It should help.