Lending is primary business for banks and other lending organizations. Banks use credit scoring to accept or reject a loan request. Rating organizations like CIBIL, Experian and others also provide individual’s credit score to lending organizations. The score provided by rating organizations along with many other parameters are used to decide a good or bad loan borrower by lending organizations.
Traditionally, lending organization’s credit score assessment was based on pre-defined parameters with weightages. Over a period of time, it became a result of a statistical model to identify good or bad potential borrowers. Recently, machine learning models are being leveraged to assess credit score due to their ability to learn from large number of parameters (features) and faster decision making. Lately, the importance of machine learning models explainability has become imminent. This is more important as any wrong decision making can lead to loss of business and penalties by regulators.
As part of this hackathon, we need you to build a CredScore software application that gives flexibility to loan administrator to define parameters (features) for any loan product (personal loan, vehicle loan, etc) with weightages. These features can be numeric, text and date data types. The values can be categorical – ordinal and nominal. This application is to be deployed as an API for modularity, scalability and management of user roles.
Loan officer can access loan applicant’s data of a particular loan product that he/she has access and perform rule and machine learning based assessments. Loan officer may also seek explainability of the recommendation made by machine learning to make the final decision. The decision of loan officer to accept/reject loan request is provided as a feedback to Credscore so that training data for ML model is enhanced.
This application helps bank official to predict the risk for the customer while giving the loans