Loans are an extremely powerful financial instrument; on one hand, they have the potential to create opportunities and foster growth, yet on the other hand they can create crippling debt, trillion dollar bubbles, and devastate the very economies they were meant to serve. If only there was a way to be able to accurately predict whether a loan will get paid back......
What it does
Predicts the likelihood of a person charging off or defaulting on a loan.
How we built it
Using the Club Loan data set, with a standard Jupyter and Sklearn workflow and a Google Compute back-end.
Challenges we ran into
As this is our first attempt at working with big data, we had a big question; what exactly is our question ? What are we trying to solve here? What are we trying to prove? Simply reaching that step of having an end-goal in mind took much longer than anticipated. So did training the models when we eventually committed to a path, despite doing so on the cloud.
Accomplishments that we're proud of
We were able to internalize and present the data in a meaningful way, beginning with exploring what factors effect the biggest culprit in loan defaults, interest rate, then being able to build off of that into creating a fairly accurate predictive model, while still getting a good night's sleep.
What I learned
Just because your model should theoretically be the most suitable for your data, doesn't mean that it is. Pre-processing makes a much greater impact on the quality of predictions than fine-tuning your parameters. Nothing impairs cognitive skills like hunger.
What's next for LDEP ?
Fine tuning our current models, as well as acquiring more detailed data sets or merging other socioeconomic and geographic data to make our predictions even better. And who knows, maybe even a UI in time...