Simple behavioral clustering using Credit Card transactions

Learn Data Science
Teradata Employee

Last summer we delivered a POV in Europe for maybe the oldest bank in the world (who recently bought Aster by the way). We had to deal with credit card transactions data and one of the objectives was to use our creativity and show some of Aster's capabilities.

Well, sure, no problem.

I have transactions linked to each customer and point of sale. After some data preparation (using the power of Aster SQL functions by the way) we have previously been able to categorize all points of sale and link them to a merchant category hierarchy.

I quickly had an idea, that might quick and simple to achieve with Aster :

Try to analyze shopping behavior similarities between customers using a «retail-like» approach and segment the customers based on their spends in different categories of merchants.

Let's consider and build a monthly basket of credit card spends. Why monthly ? For several reasons, first frequency of use of the credit card, second monthly bills and third well obviously most of the time people get paid monthly and budget spends monthly.

Here is workflow :

1. Use NPATH to create customers' monthly baskets

2. Use TF-IDF to weight each individual purchase category

3. Use LDA to identify purchase topics and that way segment customers based on their credit card spending behavior


We started the topic detection with a number of 5 which actually gave straight meaningful results.


So, pretty simple technique here, yet effective, to start playing with credit card transactions.