1
$\begingroup$

Most of the articles on the internet shows the Market Basket Analysis using data from supermarket (i.e. clients uses to buy a lot of items in one day in a supermarket), so I think makes sense to group the transactions by day.

In my work I'm doing a Market Basket Analysis, but as the majority of the clients just buy 2 or 3 items by day, I grouped the data by month (also tried to group by week). In total, I have 18k items bought.
That makes sense?

I did this to have more bigger rules, like $\{A\} \rightarrow \{B\}$, or $\{A, B\} \rightarrow \{C\}$.

Obs.: I'm using R (library arules).

$\endgroup$
4
  • $\begingroup$ I guess it depends on the purpose of the analysis, can you tell us that? For example, if you want to know what clients buy together when they are a trip to the shop, for purpose of layout of the wares in the shop, maybe it does not make sense. Otherwise, it could make sense. $\endgroup$ – kjetil b halvorsen Aug 17 '20 at 2:46
  • 1
    $\begingroup$ Our clients buy courses (they are interested to know about investments). As we have some different products, suppose that a client called Fernando bought the item A. So Fernando is gonna to study in the next days, and that's why I think he can buy another of our products, because he is engaged to get more and more knowledge. The rules that I want to get are: If Fernando bought item A, what's the most likely item he can buy? This makes sense? As we don't have too much data, I decided to aggregate this transactions by week. $\endgroup$ – igorkf Aug 17 '20 at 3:02
  • $\begingroup$ Can you please adf this extra info así an edit to the questions? Not everybody reafs comments $\endgroup$ – kjetil b halvorsen Aug 17 '20 at 3:34
  • $\begingroup$ In that case grouping makes sense. $\endgroup$ – kjetil b halvorsen Aug 18 '20 at 3:19
1
$\begingroup$

I think sequential pattern mining might be more appropriate for your use case, so you can recommend the next most likely item in sequence. Otherwise perhaps use all the customer's purchases, regardless of time. Data over a finite time period can make sense, but in your case it seems the cut off might be arbitrary. If you want to go down that route I'd do more exploratory data analysis to see what time period would make sense

$\endgroup$
1
  • $\begingroup$ Thans to cite the sequential model. This makes sense to me. I found that MBA has a variaton to be modeled as a sequential model (adding a time series component). $\endgroup$ – igorkf Aug 17 '20 at 10:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.