OneScore

Home page
Applicant comparison page
Model overview page

Inspiration

In today's economic climate we realized millions of people around the country lack sufficient credit history despite having stable financial activity. This stopped them from receiving significant credit lines despite having the finances to back it.

What it does

OneScore calculates a score for applicants based on their economic activity and financial history. Based on their spending habits, income and other parameters we determine if someone is eligible for a credit line despite their actual credit score.

How we built it

We initially created a system based on python and fast API, then created a logistic regression model with scikit learn to categorize people. Using AI we synthesized a 600 customer list of year long transactions to then train the model. Then we created a frontend and hosted the website using Render.

Challenges we ran into

Using the Nessie API initially proved to be challenging, synthesizing the data to include noise but still remain realistic was also difficult. Finally deploying and hosting the system was our last hurdle before we were able to complete it.

Accomplishments that we're proud of

Our model has a 77.3% accuracy on a 6-fold cross validation set of 600 people. Our UI/UX is also interactive and very intuitive while remaining aesthetically pleasing. We also designed a simple logistic regression model which was easy to train and still effective.

What we learned

We learned to better communicate while developing and ensure that everyone was on the same page while making our solution. Logistic regression was also new to us so we had to learn about creating models.

What's next for OneScore

If we are able to use it in practice we would be able to train the model on actual customer bases with real values with would make it significantly more accurate. It would also go through legal review before being used in a real notice.