Investigate LHD (least hit density) eviction algorithm

Least hit density (LHD) is an eviction policy based on
hit density. LHD monitors objects online and uses conditional
probability to predict their likely behavior. LHD
draws on many different object features (e.g., age, frequency,
application id, and size), and easily supports
others. Dynamic ranking enables LHD to adapt its eviction
strategy to different application workloads over time
without any hand tuning. For example, on a certain
workload, LHD may initially approximate LRU, then
switch to most recently used (MRU), least frequently
used (LFU), or a combination thereof.

http://www.cs.cmu.edu/~beckmann/publications/papers/2018.nsdi.lhd.pdf