This relatively simple question collapses neatly into the title. I have large swaths of numerical and categorical data that contain unknown amounts of signal relative to weak data modeling principles. From these data sets, I'm attempting to extract usable signal -- that is, data that is well-ordered, correctly transformed and modelled, and useable for stochastic ensemble construction and analysis that occurs downstream.
In order to do this, I have decided to construct what I am calling a Shannon Filter: a simple, naive system that computes the Shannon information entropy of each piece of data with respect to its current model, ranks the data in descending order (relative to a logitnormal distribution), and performs filtering at some confidence level yet to be determined.
While this is not yet a methods question (I may turn to http://dsp.stackexchange.com or http://crypto.stackexchange.com for that answer, since both spaces contain direct analogs to my problem), it is currently one of correct terminology. The phrase "Shannon Filter" appears to be in loose current usage, yet I can find no sourcing to back this up.
Would use of this term be considered appropriate in the statistical domain?
I ask here, because of all the StackExchange sites that delve into this class of problem (inclusive of math, dsp, crypto, cs-theory, SO, and english; yikes!), abstract statistics seems to be the most relevant domain that I wish to apply this term within. I expect my system to be defined in the context of stochastic sampling and pairwise search algorithms digestible by statisticians, so I require a crisp, clear definition for this process that is technically correct.
Thank you in advance.
