Inspiration

My country lacked an individual credit score system to help financial institutions. The manual method of predicting defaulters wasn't really fair. So I wanted to bring a change to that.

What it does

Given a list of financial information of an individual the model assigns a score to the individual and shows proper visuals and indications to why the score was assigned and which features affected the score positively and negatively. Providing clear interpretation of the model.

How I built it

I built and optismied the model in jupyter notebook in python. Using popular ml frameworks like numpy, sklearn and pytorch.

Challenges I ran into

The main challenge was to find a reliable dataset to work on and to provide proper interpretetibily of black box models.

Accomplishments that I'm proud of

I'm proud of the results from my model. It provides clear interpretation of the score assigned. And on testing I found out that the interpretations mimic the human evaluation process. Based purely on the financial data, and not any other racial or cultural biases.

What I learned

I learned core concepts of how different models work especially Random Forests, Gradient boosted trees, Neural networks. Interpretation of black box models.

Built With

Share this project:
×

Updates