The problem
Many creditworthy people have no formal credit history, but models trained on thin data can encode bias.
What to build
Design an explainable credit scoring prototype using consented alternative data and a documented fairness analysis across groups.
Success looks like
Comparable approval accuracy across tested groups, reasons given for every decision and an appeal process in the design.
Suggested technologies
- Explainable ML
- Fairness metrics
- Alternative data
- Model governance
- Synthetic data
An original, independent statement. It is not an official Smart India Hackathon problem statement, and Arventiq Labs is not affiliated with or endorsed by that programme. The scenario is illustrative and does not describe any specific organisation. Free to use with credit.