A key characteristic of Bloom Filter is that it only generates false positives but does not have any false negatives. However, in many applications such network monitoring, besides accelerating the membership checking procedure, we should also ensure that there are not any false negatives.
Thus, how to effectively reduce the false positive rate of Bloom Filter without generating any false negatives?
I tried to search the raw dataset after using Bloom Filter for approximate membership checking, but this procedure causes large computational costs and uses extra memory space to store the raw dataset.
There are a few ways to reduce the false positive rate. First, you can ensure you're using the optimal number of hash functions. Check the Wikipedia page on Bloom filters to see what that is. Second, you can increase the size of the Bloom filter. Third, you can use a more space-efficient filter like a ribbon filter.