Hi @JvWyatt,
If you search the forum for "mixed orientation taxonomy" you'll see that the most common reason for obtaining poor taxonomic assignment is due to the fact that the reads are oriented in the opposite direction compared to the reference databases. That is, when using feature-classifier classify-sklearn the reads must be in the same orientation as the reference database.
As it appears that all of your reads are in the wrong orientation, you can follow the 2nd recommendation here, and use RESCRIPt to re-orient the FASTA file output from DADA2.
Alternatively, you you can re-orient your fastqs by flipping how you import your R1 and R2 reads. That is, import your R2 reads as your forward reads and your R1 reads as your reverse reads. Then re-run DADA2 and then your taxonomy assignment. This will have the effect of reverse complimenting your reads, putting them in the 'correct' orientation for the classifier.
One quick test you can do, prior to running the above two approaches, just to sanity-check your data. That is run feature-classifier classify-consensus-vsearch. This method does not care about sequence orientation, and should provide reasonable taxonomy assignment. But be wary of making phylogenies or comparing this sequence data other projects, as the sequences would appear very different due to differences in orientation.