As per your suggestions, I ran qiime feature-classifier extract-reads command-line. My NCBI_REF sequence got reduced from 21,624 to 16 sequences after the extraction.
This was the command line I ran
I ran this command-line multiple times, by changing value trunc length --p min and --p max length, but my sequence count got reduced even more bad. It came three and five sequence finally.
What should I do further?