I am working through the Adventure Works data mining examples on the Microsoft website. In it, we are going to train a model using all sales data globally, then use the data for a region and bike model as inputs. Wouldn't this just predict incorrectly, ignoring specific trends for that area for that bike model?
What would be the advantage of doing this?
I think the idea that is that, when developing a learner, global data encompasses regional data. If you're building some sort of classifier and hope to run it at a regional level, you only need to use the regional-specific data, no?
Every model needs to be trained with relevant data.
The confusing part is that perhaps I'm not understanding the differential of "regional" data. Ultimately, the global data should definitely be relevant to your predictive model.
