You can just grab a few reads at random and do an alignment to the reference database sequences to see if they align in the complement or reverse complement orientation.
Not in QIIME 2, but this looks promising: OrientNucleotides: Orient Nucleotide Sequences in DECIPHER: Tools for curating, analyzing, and manipulating biological sequences
This issue crops up every now and then, so I will add a method for read orientation in a future QIIME 2 release. That will make it easier to implement in your pipeline, but for now your best bet is to export and use a tool like decipher, then re-import to QIIME 2 for classification.
See the link to the paper I provided above, which describes the parameters and their effect on classification accuracy. In general, use defaults unless if you are attempting to benchmark, e.g., on a new method or primer set, etc
I hope that helps!