Transactional patterns


Hi fellows

Is there any technique/algorithm which may be useful for finding transactional patterns in the banking sector. The requirement is to identify any kind of transactional pattern and this needs to be done at customer level. So the output needs to have transactional pattern for each customer if it exists.

Thanks !!


Use a tree algorithm and build a tree so you can look at some patters. Let the tree grow to the fullest as your intention is not to predict anything.

This is an indirect approach to quickly figure out some repeating or separating pattern between different transactions. I may be wrong, but nothing wrong in trying it out.


Hi Vivek

In this scenario we don’t have any target variable and this scenario falls under unsupervised learning. The objective is to find the transactional patterns at customer level.
Some examples of patterns i can think of are:

  1. Customer does a transaction from Account1 to Account2 every month in first 5 days for xyz amount
  2. Customer does a transaction from Account1 to Account2 in second month of each quarter for xyz amount.
  3. Customer does 60% of the debit transactions to Account2.
  4. 70% of the customer transactions are done in first 4 months of the year.
  5. July and August constitutes only 5% of the customer’s total transactions done in the year.

I know these can be done based on rule-based approach without the use any machine learning technique. But with rule based approach we need to first decide what are the patterns we are looking at and then do the coding accordingly.


interesting question. I’m eager to know how guys would go about this and what kind of data you would collect.

For example; imagine we have customer age,tenure,credit txns,debit txns and product holding. is it possible to find a pattern using this information only?

i could do a cluster analysis using this data which will only tell me customers with similar behaviors, but it doesn’t necessarily give me patters right?