0
$\begingroup$

I'm working on a model to predict a customer as being 'in-market' for a product in the next 6 months. The dataset has a wealth of information like lifestyle and demographic variables and previous transaction history. What is the best model to combine the demographic and lifestyle data with transaction history to predict a purchase in the next 6 months? Would a combination of a model that predicts likelihood to purchase with something like a survival model to predict when that purchase might occur work?

Are there any real world examples of something similar to this? Basically predicting time to purchase.

$\endgroup$
  • $\begingroup$ This is a typical problem where unintended biases can ruin the validity of your results. Does your dataset contain gender, for example? This protected variable can sometime offer great predictive power, but with great power comes great responsibility. Ensure that just because a person is labeled "male," they are not always assumed to someday buy something male-oriented. Likewise, ensure your model doesn't make offensive purchase prediction based solely upon race, zip code, income, political ideology, relationship status, sexuality, etc. Lots of work done on this if you need more guidance. $\endgroup$ – Alex L Aug 29 at 18:13
0
$\begingroup$

This is a forecasting problem, you are predicting a value in the future for a group of people which is the day they would make the next purchase. Sure, you would need the last time each person has bought something ( that's your target ) , periodicity of purchases;number of times that person bought something,obviously the id of the person as variables. And those are the things on top of my head, maybe someone who worked on forecasting problems could help you better.

On a second note : this is how you could approach the problem; maybe encode the dates they made the last purchase as date_num 1 to infinity, and each number in this target variable represents a consecutive day starting from a certain date(maybe the oldest date in your dataset). You would do a regression to try to predict that number using every potential predictor and round it to get an integer representing a unique day.

$\endgroup$

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.