Looking for the name of an algorithm

Thread Starter

Holes Flow

Joined Oct 26, 2017
14
At first, I will post this as short as I can, as the full question can be lengthy, but provide context.

I'm old, but I have this recollection of an algorithm that scores, or categorizes (somewhat) a variety of temporal inputs, giving as output those inputs that are related to one (or I guess more) input. I've combed through my ML and ANN/NN books and have come up empty. Google generates way too much noise.

A use case is a collection of 100s of sampled variables about a business, say daily. The input of interest is Sales ($)-specifically sales in the next week (for this use case). Many of the 100s of variables have no correlation to future sales, but some do. I'm looking for those. I realize the sample window is variable. A few examples:

When input X goes down, sales go up in the next week. When input Y goes up, sales go up; and input Z going up has future sales going down. You get the idea.

I'm trying to be as short as possible here, but am happy to expand if needed. My goal is to run the historical data through the algorithm (inputs and results) until the relevant inputs are sorted from the irrelevant, like weights in normal NNs. This way, I can run other more common algorithms on just the relevant inputs in future with a trained traditional algorithm.

I can't think of the name of this algorithm/system I have read about, and am grateful for someone more familiar. TIA!
 

WBahn

Joined Mar 31, 2012
30,072
I think the general term you might be looking for is 'correlation' -- to what degree does one set of data relate (i.e., corelate) to another set.
 
Top