By: Herman Jopia, First Vice President and Data Analytics Manager, American Savings Bank
One of the main concerns in a credit scoring project is the extraordinary amount of time required for its development, usually months after the data has been collected.
Most of this time is concentrated on a particular stage of the development called the “Generation of Predictive Characteristics”, a process that generates a list of features that have the potential to become part of the final model.
This list can be huge depending on the creativity of the modeling team, likely on the hundreds mark, and the analysis of each one of them is the reason why it is so time consuming.
This article discusses the usage of specialized code to dramatically decrease the time spent on generating predictive characteristics using a real life example from a Chilean Bank while developing one of its credit scoring models for account management.