It looks like you removed the PCR primer sequence from your nextera adapters and are trying to trim using the nextera adapters, which are not biological. You removed the part that you should be using here, the PCR primer. You should use:
CCTACGGGNGGCWGCAG
GACTACHVGGGTATCTAATCC
Well, yes, that is another option. Training a classifier specific to your 16S rRNA subregion will increase accuracy a little bit, but not much. The full-length 16S classifiers will still perform well for classifying subregions, so you can use that if you continue to encounter problems during training (right now the issue is that you are not inputting really primer sequences that match biological sequence. but once you get it working you could run into other barriers like memory constraints).
I hope that helps!