Time Series & Feature Engineering Questions
Hi,guys
I am trying to solve a sales forcast problem: there is a monthly sales table(attribute:time,sales) and a consumer record table(attribute:order time,A(id),B,C,D) , assuming that sales are related to the consumer's attribute, How should I creat proper feature as input to build a timeseries model to make predictions?
there are many ways to create input by counting instances in different dimensions as input. as follows
var1 =count A when B=b1,
var2 =count A when B=b2
var3 = count A when B=b1,C=c1,D=d1
var4 = count A when B=b1,C=c1,D=d1
…
How to select proper input for time series prediction from these variable?Is this the right way to creat feature?
Anybody have any ideas? Appreciate a lot for any tips! Would u mind looking at this?:p @Thomas_Ott
Answers
@rpleaner do you have a sample process to share?
I try to build simlar input tables using Generate Data Operators in Rapidminer as follows.
Supposing that customers' attributes are related to monthly sales, how to create proper input for time series forcasting of next month's sales ?
@Thomas_Ott
@rpleaner you can try something like this: