FairLend

after mitigation algo
bias detection

Inspiration

If you've ever tried to buy a house, you probably have an idea how complicated it is to land your dream home. Things like the mortgage rate you qualify for, bidding wars, and having enough credit to confidently bid against cash buyers are all par for the course.

In Canada, Black people typically face additional complexities before they even start to dream about that sweet corner lot. On top of the various criteria lenders use to figure out how much money and under what terms they might lend it, Black people are more likely to encounter white supremacy in all aspects of the housing market, meaning they are less likely to be able to rent or buy desirable property.

What it does

FairLend detects and mitigates bias in loan approval by using algorithms and fairness metrics to analyze and adjust lending practices, correcting biases that disproportionately affect marginalized groups, thereby promoting equal opportunity in the housing market.

How we built it

Generated synthetic data due to time constraints and performed a basic analysis to check for any apparent biases in loan approval rates between different races. Split data into training and test sets. Then applied Logistic Regression because this is a binary classification problem; a loan is approved or disapproved. For the mitigation strategies:

Adjusting Decision Thresholds: The decision threshold is adjusted based on the Receiver Operating Characteristic (ROC) curve to improve model fairness.
Re-weighting Training Data: The training data is re-weighted to give more importance to underrepresented classes, aiming to balance the model's performance across different groups.

Challenges we ran into

Started off with analyzing data, pivoted after realizing I was focusing on the wrong thing. Instead, I generated synthetic data.

Accomplishments that we're proud of

The code working in the end. This is the first time I have conceptualized something and finished it within the weekend. Looking forward to developing it further.