Digital footprints in credit scoring
summary
E‑commerce logs plus a private credit bureau score train a plain logistic regression using 10 digital-footprint variables (device, email type, typing errors, referral channel, grammar, etc.). Digital footprints alone deliver an AUC that matches bureau models, and a combined model improves AUC further; correlation between datasets is ~10%. Footprints rescue 6% unscorable customers and surface intuitive proxies (iPhone → lower default, paid email → lower risk, numbers in email → fraud), with clear regulatory and behavioral caveats.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 3:00Credit scoring challenges
- 5:51AI regulations in banking
- 6:53Digital footprint data
- 9:46Digital footprint study
- 15:04Model performance results
- 17:07Impact of digital footprints
- 20:02Companies using footprints
- 21:04Customer behavior insights
- 22:30Regulatory and ethical issues
- 24:17Bank and fintech opportunities
- 25:19Q&A
related talks
06mWill Online Digital Footprints Reveal Your Relationship Status? An Empirical Study of Social ...
52mModern Fraud Prevention Using Deep Learning
31mData Science for Financial Inclusion
39mMachine Learning in Financial Credit Risk Assessment
21mMaking Predictions Under Lending Regulations
16mAI-Driven Behavioral Analytics for Real-Time Fraud Prevention
01h 13mGoverning credit scoring. Data protection, algorithms & surveillance.
58mNew-type Scoring Models: Analyzing Customer Actions