Changepoint Detection using Store Level Sales Data


I’m very new to bayesian modeling and am trying to figure out if the idea I have for how to model a particular dataset makes sense and/or is feasible.

Apologies for being a bit vague about the industry, but the data I have is monthly transactions down to the product level for a large number of different products and companies. I have the transaction data at the facility level. I’m interested in doing some sort of online changepoint detection, where we could use the data to identify shifts in the purchasing pattern that might indicate that the growth rate in sales for a particular product is changing significantly.

At the facility level the data is pretty sparse and noisy, but my thought was that you might be able to detect a change in spending patterns if it is consistent across a large set of facilities. Does this general approach make any sense or am I looking in the wrong direction?


Hello, mps. So you want to see if there’s some universal changepoint across timeseries? This seems like a valid statistical question, but one I would not know how to do, since it involves timeseries (something which currently takes more effort to get to work in Pymc3). To get a start on changepoint detection, you could look at the thread I started once, trying to go over how to do it (and receiving absolutely excellent advice from the community).

I am sure there are some people in this forum who could give you the advanced advice you need once you’ve done basic changepoint detection.

1 Like