Conditional Weight of Evidence for Gradient-Boosted Credit Models: Estimation, Additivity and Its Relation to Adverse Action Reasons
Traditional weight of evidence assigns a credit characteristic a value based on its association with default across a population.
ProofPaper ↗
Key points
- We study conditional weight of evidence for this purpose.
- We develop estimators using gradient-boosted trees and evaluate their accuracy in a bureau-style simulation.
- An application to FICO HELOC data illustrates estimation when the population evidence is unknown.
- Conditional weight of evidence provides an applicant-specific interpretation of the information in a credit characteristic; its population summary, a conditional information value, splits the joint information in the record across blocks when averaged over orderings.
Sources (1)
- [1]Conditional Weight of Evidence for Gradient-Boosted Credit Models: Estimation, Additivity and Its Relation to Adverse Action ReasonsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 17, 09:47 AM
Traditional weight of evidence assigns a credit characteristic a value based on its association with default across a population.
We study conditional weight of evidence for this purpose.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 16, 2026Continuous accountability for Verified bots
- Oct 15, 2026Shortcut Sensitivity in Agent-Safety Evaluations
- Oct 15, 2026Friends don’t let friends feed their data to AI agents
- Oct 15, 2026Cloudflare and OpenAI: making AI discovery better for everyone
- Oct 15, 2026Each Reward to Its Relevant Tokens: Reward-Specific Token-Level Credit Assignment for Multi-Reward RL
- Oct 15, 2026ManiFusion: Unlocking the High-Throughput Generation via Superposition in Manifold Space