Getting a loan depends on the customer’s reliability, or the bank’s desire to give a loan with the possibility of a default. Banks lose significant money from giving loans to customers who will not pay up and lose potential clients from too stringent loan policies. Evaluating customer reliability through improving loan default predictors can vastly aid in maximizing bank profit.

Smart Loans teams with Capital One and uses top of the line ML algorithms to improve pre-existing loan default predictors. WIth Smart Loans easy user interface, customers who would like know if they qualify for a loan can quickly rely on our platform.

What it does

The user (customer) inputs relevant information: Loan amount e.g. 10,000 (USD), Monthly payment e.g. 300 (USD), Bank balance e.g. 5,000 (USD), and Credit score e.g. 630. We output if the customer will default, if it's a risky loan, or if it's safe to give out the loan to the customer.

How we built it

We use 4 different ML Regressions (Logistic Regression, K-Neighbors Classifier, Decision Tree Classifier, Linear Discriminant Analysis, SVM) to predict chance of loan default. Our model is tested and trained on Lending Club data (with 77% accuracy, 0.001 SD) using the LR model.

Challenges we ran into

We encountered a few challenges analyzing the dataset from the Lending club, specifically extracting data and dealing with missing data points. In addition, we had little ML knowledge so we learned about different machine learning models: Logistic Regression, Support Vector Machines, etc. Lastly, 3 of us never worked with APIs, so working with Capital One's API was challenging at first.

Accomplishments that we're proud of

We created an ML Model with 77% Prediction Accuracy and an aesthetic, simple Sign-in Platform for customers. Although we came in with very different backgrounds in UI, API and ML-knowledge, Python, JavaScript, and React, we are proud of splitting up the work evenly and communicating effectively as a team.

What we learned

We learned many ML techniques, how to code in Python (Pandas), UI design, and how to work with APIs.

What's next for Smart Loans

We would like to improving predictive algorithms to be more accurate, specifically integrate more consumer features, have more dynamically modelling and updating more constantly, and possibly even automating loans. Additional models could assist banks in setting interest rates and terms of loan agreements, evaluate the overall risk on a bank's portfolio, and alert borrowers and creditors if the risk of default on an existing loan increases.

Share this project: