Algorithms should inform book culling decisions made by the Librarian, but they shouldn't replace the Librarian's knowledge and expertise.
How can circulation numbers take into account people who aren't using the library because it already doesn't have anything they want? I can't be alone in this, but simply looking at circulation numbers won't capture that, and piloting with a book or two from those categories won't get me back in (I probably won't notice because, after many disappointments, I've mostly stopped looking). Let an algorithm go nuts on that data without human judgement and I suspect you'll gradually put more and more people in my position, unless it's a damned smart one.
1. Regular strategic planning which includes a community survey. This is conducted by an outside firm and non-users are targeted as part of the survey. We also do focus groups.
2. Interlibrary loan. You local library can likely get titles from other libraries. This should inform the collection.
3. Patron driven acquisition. Some way of suggesting books should be available. We buy most suggestions and find that books that are suggested circ better than ones selected by staff (generally, anyway).