My tool "scorecardbuilder" is a tool to develop credit risk scorecards. Risk scorecards are used by banks to determine whether a customer is credit-worthy to be granted a loan, as well as the pricing of that loan. An example of such a scorecard would be the FICO score, very popular in the United States as well as some other countries.
The submitted application comprises
- An interface for end-to-end development of a scorecard starting from data upload to scorecard validation (since such data is not easily accessible, the user has the option to download simulated data from the tool and use it)
- An extensive help system to guide anyone with some knowledge of linear models to learn how to build one (every screen has a Help section with instructions; clicking on the ? icon at the top right provides some additional details and terminology)
A few technical details
- Shiny modules - this app makes extensive use of the module system in Shiny (at the time I was starting the project, Eric Nantz's talk in the RStudio conference on passing reactives between modules had just been published online and was a great help)
- rhandsontable for editable tables
- smbinning for the conditional inference tree algorithm of binning variables if the user wishes to use it
- a subset of the tidyverse packages for data manipulation
Links
Github (codes/libs.R lists all the required packages): GitHub - anindyamozumdar/scorecardbuilder: R/Shiny interface to build a credit risk scorecard
RStudio Cloud: Posit Cloud
Shinyapps: https://radmuzom.shinyapps.io/scorecardbuilder/
Enhancements and Improvements
While the app is complete for the purpose of this contest, it requires much more work before it can be used in production. The Github page lists a number of enhancements and I will work on them over time. Also, the code requires quite a lot of restructuring and refactoring - lot of the functionality are codes inside Shiny render functions which should ideally be separated into their own logical groupings. No work has also been done on optimization and efficiency. If anyone wishes to collaborate, please contact me separately.