The best thing to do is a phylogenetic tree in my opinion.
The trendy approach and the one deemed acceptable in publication is deep learning for precisely this application, such as the pipeline published here. This approach is attractive in 16S NGS metagenomics classification because it uses advanced approaches in deep learning, notably convolution neural networks (CNN) and deep belief networks (DBN). I don't really understand DBN, but I suspect it is a variant of recurrent neural networks (RNN) such as long-short term memory (LSTM).
Having said all that the accuracy is low, viz.
For instance, at the genus level, both CNN and DBN reached 91.3% of
accuracy with AMP short-reads, whereas RDP classifier obtained 83.8%
with the same data.
91% accuarcy for a deep learning neural network is minimal, but it will be quick and will do precisely what you want with little fuss. A phylogenetic tree on the otherhand can take several hundred hours of total processor time. The thing about a 16S tree is that it is 100% accurate, because you can manually identify "grey zones" to assign their classification.
You can retrain CNN models and in future this will likely resolve the accuracy issue. However, I don't know about about DBN and I don't think RSS can retrained, more specifically I don't know how to do it.