Credit risk is the risk of default on debt obligations by the borrowers. It occurs when the borrowers fail to meet their contractual debt obligations and are unable to repay the borrowed amount of money during a specified period. Fineract credit scorecard is an AI powered solution that allows users to perform credit assessment of loans using various techniques such as rule based based scoring, statistical scoring and machine learning scoring.
Financial institutions have been struggling for years to make sure that they minimize credit risks. Credit risk can lead to heavy losses for money lenders like banks and other financial institutions who lend funds to borrowers for a specified interest rate. Credit risk can disrupt cash flows for the lending party. It arises when money is not repaid within a specific period as agreed upon earlier. The risk and loses for lenders includes lost interest and principal amount. It can lead to an increased cost of collection for lenders.
In lights of this the Mifos Initiative has been trying to put in place a credit assessment platform into their financial inclusion platform. This integration is mean to improve the scope of the Fineract platform by empowering financial institutions with the ability to use mathematical methods confidentially assess the credit worthiness of their clients.
In this years GSoC, the main goal was to refactor, improve and deploy the work that was made in the previous years. A couple of planning and steps were made to bring this integration to a usable state.
-
In the past iterations of this effort, a lot research and investigations were done to try to fine the suited approaches and solutions to the credit risk assessment of financial institution. This work was documented, some models were developed and tested.
All this efforts needed to be refactored, cleaned up and a RESTful API built around it so it can be used in real world scenarios. For that reason, the work was ported into django application with django rest framework. This REST server comes with the ability to work independently as a unit on it's and deliver various credit scoring methods. This server comes with the ability to track and keep record credit assessment operations for future use and references so the model can be self improved over time.
-
The next step for this effort was to integrate the solution into the Fineract platform as a reference implementation. This involved developing both Backend and Frontend integrations in the Mifos Web-App and Fineract 1.x backend so the solution can be used realistically in a production scenario.
-
Another important component of the work was to provide a testing framework for the modules and integration; this involved both Unit and Integration tests. These tests were written and integrated in all the components and implementations.
-
The final piece was the documentation for the project about what it is and how to use it.
Fineract users have been lacking a proper way of assess the credit worthiness of their clients for years and during this summers Google Summer of Code, extra focus was set on finally making the previous years of investigation and research on the topic available to the public.
The credit scoring engine for Fineract took a step forward during this summer, delivering multiple scoring capabilities methods and algorithms and integrating a reference implementation within Fineract 1.x and the Mifos Web-App.
Though a lot was delivered during this summer of code session, the credit risk assessment engine could still use some improvements. First off will be improving on the engine using H2O.ai which could largely improve the accuracy and precision of the scoring engine. Next would be developing an ML pipeline so the ML scorecard engine can be improved with time.
