CALD Seminar

  • Richard Scheines & Peter Spirtes

The Mellon Blitz Project: Predicting and Causing Checking Account Attrition

We were given a Mellon dataset on approximately 90,000 checking accounts which contained information about the account activity up to 6 months before the datacollection moment, demographic information about the account holder (e.g. age). and a binary outcome variable: openclose which was oprn for over 99.2% of the accounts. Out task was twofold: one, could we find a model to predict which accounts were about to close, and two, could we identify causes of checking account closing. For the first question, we used a number of commercial data mining packages and did quite well. For the second, we used a new search algorithm that we will describe and report on results.
For More Information, Please Contact: 
Catherine Copetas, copetas@cs.cmu.edu