Gradient boosting — XGBoost wins the leaderboard and hides the reason
Boosted trees are the default winner on tabular data, but their accuracy comes wrapped in opacity — a strong model whose decisions need a second tool (SHAP) just to be explained.
Not yet scoped.
Why
The PoC pairs an XGBoost model with SHAP attributions on a credit-style dataset, and treats the gap between accuracy and explainability as the real deliverable.
How it works
Not yet built.
Where it lands in Jayverse
Number: pair any gradient-boosted indicator with SHAP attributions before publishing the reading, so admin users see which inputs drove a score instead of just the number.
Auditor: record which explainability method ran alongside any ML-derived indicator or risk score used in Verex or Number, and treat that as a checkable field.
Key expressions
Words and phrases from this page worth keeping, with the Korean meaning and the sentence they come from.
Expression
뜻 · 쓰이는 자리
leaderboard
리더보드, 순위표 · 여러 모델의 성능을 비교해 순위를 매긴 표. "XGBoost wins the leaderboard and hides the reason"
wrapped in
~에 싸여 있다, 동반되다 · 좋은 성능이 어떤 단점을 함께 데려올 때. "their accuracy comes wrapped in opacity"
opacity
불투명성, 설명하기 어려운 상태 · 모델이 왜 그렇게 판단했는지 알기 힘들 때. "comes wrapped in opacity"
the real deliverable
진짜 산출물, 실제로 목표해야 할 결과물 · 겉으로 보이는 것 말고 핵심 성과. "treats the gap... as the real deliverable"
credit-style dataset
신용평가 유형의 데이터셋 · 대출 심사처럼 사람에 대한 판단이 걸린 데이터. "on a credit-style dataset"
XGBoost
익스트림 그래디언트 부스팅(Extreme Gradient Boosting) · 정형 데이터에서 기본 우승 모델로 꼽히는 트리 부스팅 라이브러리. "XGBoost wins the leaderboard and hides the reason"
SHAP
섀플리 값 기반 설명 기법(SHapley Additive exPlanations) · 블랙박스 모델의 예측을 변수별 기여도로 분해해 설명하는 도구. "a second tool (SHAP) just to be explained"