UnityLoan

Inspiration

For underrepresented communities, the practice of redlining has historically led to unjust denials of loans based on their race, ethnicity, or the neighborhood they live in. Even though redlining is illegal, its subtle effects can still persist, affecting communities' access to fair financial opportunities. Our project aims to develop a machine learning model that can detect instances where a loan should have been approved based on the applicant's qualifications but was unfairly denied, shedding light on potential redlining practices.

What it does

The program takes certain user profile features such as income, type of loan, etc... The program outputs whether the loan should have been approved or not.

How we built it

With the user input, we run the vector through a logistic regression model that has been trained on loan data with the same features and labels for whether the person got the loan. We used certain performance metrics such as accuracy and ROC to visualize how the model works. We also found the optimal threshold for the classifiers to make decisions.

Challenges we ran into

Runtime for final product was too long, based on unzipping the really big data source and also running the model takes a while. Cleaning the data was also very time consuming, as well as doing some feature engineering.

Accomplishments that we're proud of

The user interface is minimal but conveys all the information necessary, we managed to build a linear regression model from scratch. We also believe, that if implemented to scale and more data, our design would be able to change the lives of many marginalized populations and make them aware of when they have faced injustices.

What we learned

The most difficult part of this process was managing our coding environment and being able to find and clean a dataset that is able to fit our needs. Once you have that, the project can go much smoother.

What's next for UnityLoan

We want to add visualization to pinpoint what features a user could have that may be related to discriminatory practices (if refused the loan). We want to also provide more resources to learn about redlining and how to take action against it. We also want to get a bigger data set with more user features such as disabilities, sexual orientation, and any other features that might unfairly impact the approval of a loan.