0
$\begingroup$

I am working on a Retail Company's in-store transactions for 3 months. I have performed the Market Basket Analysis on the same and I'm getting hundreds if not thousands of association rules. I am using the apriori algorithm from mlxtend.frequent_patterns import apriori in Python and I have used different support values in apriori(basket_sets, min_support=0.01, use_colnames=True), all the way from 0.01 to 0.4.

If I use a support value too high, (for some stores there are no rules found), there are very few association rules, and if I use a support value too low, its very difficult to make sense of the association rules generated since there are too many of such rules.

Since I've chosen to go with the low support value, I wanted to understand ways in which I could present the rules which would make business sense to the data owners. If there is any literature (I've tried googling at least 10 different queries but all of them return "How to do Market Basket Analysis and its applications!") on how to present the Association results, that would be really good.

Thanks

$\endgroup$
3
$\begingroup$

The best way to understand multiple association rules is to visualize them. This makes it even easier to present. This paper covers multiple approaches for visualizing association rules. Go through its references. They also suggest their tool, but it is in R. If you want resources for python try searching for "association rules visualization python" and you'll find some resources.

$\endgroup$
1
  • $\begingroup$ Thanks a lot for this answer. I not marking this as the answer yet since I'm waiting for more perspectives if any. $\endgroup$ – Pushkaraj Joshi Apr 21 '20 at 9:31
0
$\begingroup$

The paper linked by bkshi looks like an excellent resource for visualization. I don't know if you considered any metrics other than support, but filtering on confidence and lift is also supported by the mlxtend package and might also help you narrow down your rules in terms of value to the business.

$\endgroup$
3
  • $\begingroup$ The issue isn't with selecting the best-suited rules, its more to do with presenting the results and convincing the stakeholders with your efforts. $\endgroup$ – Pushkaraj Joshi Jun 10 '20 at 16:54
  • $\begingroup$ Yes I understand that and the previous answer has a great reference for visuals. But in addition, since you said "its very difficult to make sense of the association rules generated since there are too many of such rules" and you therefore want to visualize them, it makes sense to consider the other metrics to help narrow down before visualizing. A rule with even very high support but low lift or confidence is likely not valuable to a stakeholder anyway, so if you're not considering those additional metrics you will be showing stakeholders a lot of noise regardless of the technique you choose. $\endgroup$ – elz Jun 10 '20 at 19:37
  • $\begingroup$ I understand what you are saying and thanks for your replies. In the market basket analysis, even after filtering upon the criteria that you have mentioned, there are simply too many rules. Hence the problem becomes how to visualize those rules and make those "many many" rules presentable. $\endgroup$ – Pushkaraj Joshi Jun 11 '20 at 4:55

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.