10

To build models to predict customer behavior, I am searching for transactional data over multiple years (i.e. > 3 years). My focus is to assess the quality of long-term predictions, thus the longer the time period the better.

Minimum requirement in terms of variables would be:

  1. Customer ID
  2. Purchase/order/transaction date
  3. Price of goods/services sold

In a perfect world, these datasets would be released under a GPL license and also include information like:

  • Date of first purchase ever (which facilitates cohort analysis)
  • Socio-demographic information on the individual customer
  • Product ID, as well as further details of the products bought such as product category

Examples are manifold and include order/transaction/purchase histories from an online store, a car/bike share network, transactions from a gas station network, ...

I do know about the Superstore dataset that ships with Tableau and has 4 years of transactional data (http://community.tableausoftware.com/docs/DOC-1236). However, this is simulated data and does not show the patterns one usually observes in real world transactional data.

  • Any luck finding any good transactional data like this? Amazing how hard it is to find.. – dashnick May 22 '17 at 23:34
  • Unfortunately not. – majom May 23 '17 at 6:33
2

Kaggle once conducted a competition with 22GB of real transaction data:

But look after the terms of use (only for the purposes of competition) and so on.

  • I am aware of this dataset, but it is only covers transactions for 1 year. I am search specifically for a dataset > 3 years. – majom Dec 20 '14 at 9:21
  • 2
    By the way, there are many "short" transactional datasets available, see e.g. marketingedge.org/marketing-programs/data-set-library. – majom Dec 21 '14 at 11:41
1

Sites that aren't "stores" are more likely to share customer data. So take a look at the crowdfunding sites:

You can get a set of contributors by looping over campaigns.


If you are more ambitious, you could do the same for an online store. You'll have to do some research to see if any fits your needs. I think with a combined API and scraping approach, you can build a decent dataset of customer transactions.

Each transaction can have up to two feedback records, one left by the buyer for the seller, and one left by the seller for the buyer. Feedback is optional; so not all transaction will have feedback records (although most will.)


Regarding the demographics, I doubt any site will include demographic information - but consider predicting age and gender from name).

  • Can you really scrape the individual amount that any person has pleged to a Kickstarter/Indiegogo campaign? I already did scrape Kickstarter quite some time ago and at that time this information was not accessible. – majom Dec 21 '14 at 11:39

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.