Panda Trust

Inspiration

We were inspired by the prevalence of Bakong, a blockchain-based p2p payment service in Cambodia where majority of the population remains unbanked but owns a smart phone. We thought that such a demographic that is shared by other developing regions provides the means for a p2p lending platform among other open banking services to prosper, but the inherent issues of trust and information asymmetry in p2p lending marketplaces has led to failure in implementation.

What it does

Our API will predict a borrower's likelihood of default using data gathered by the lending platform, such as during the registration and KYC process.

How we built it

We cleaned up the Bondora peer-to-peer lending dataset found on Kaggle, which we then used to train our XGBoost model with tuned hyper-parameters.

Challenges we ran into

Initially, we wanted to base our model on NLP but realised we didn't have the capabilities to do so.

Accomplishments that we're proud of

That we came up with something haha.

What we learned

Data handling is hard.

What's next for Panda Trust

We would love to explore using NLP for predicting bad debt, and perhaps explore having a mix of "hard" and "soft" indicators for our prediction.

Built With

• python