prasunshrestha / loan_default_prediction

Our objective, therefore, in this project, is to build a classification model that will predict whether a borrower will default or not. We have used Lending Club’s loan data from 2007-2015, which includes pertinent payment and borrower’s information.

Loan Default Prediction

Peer-to-peer (P2P) lending has transformed how people have access to credit and investing, as it connects borrowers and lenders, both of whom are individuals. Whether it be debt payment, small business growth, or investment in the future, these individuals, who might not qualify for lending practices by stereotypical measures can now participate in this ecosystem and help each other.

Lending Club is a pioneer in the P2P industry. Since 2007, it has over 26 billion dollars in loans to more than 1.5 million borrowers. Lending Club verifies if both investors and borrowers meet their criteria for a transaction. However, P2P lending poses a substantial risk for the lenders because the deposits are not secured, or FDIC insured. As a result, the lending practice can result in a loss for investors if a borrower defaults.

Our objective, therefore, in this project, is to build a classification model that will predict whether a borrower will default or not. We have used Lending Club’s loan data from 2007-2015, which includes pertinent payment and borrower’s information. The original dataset lending club is a matrix of 2260668 observations and 145 features. We build an ensemble of classifiers - Random Forest, Logistic Regression, and Linear Discriminant Analysis (LDA) - and our model outperformed the baseline model by 40%. Finally, we catered the decision boundary in our model based on an investor’s risk appetite.

Note: The files in this repository are uploaded in retrospection.

Date: Spring 2020 | Collaborators: Varsha Jain, Radhika Khandelwal, Akshay Tatyarao Munde, Prasun Shrestha