Skip to content
View junhyung-L's full-sized avatar

Block or report junhyung-L

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
junhyung-L/README.md

이준형 Junhyung Lee | Data Scientist

고객·정책·공간 데이터로 예측 모델과 의사결정 시뮬레이션을 만들고, 각 결과가 어디까지 유효한지 함께 밝힙니다.

  • 현재 하나금융융합기술원 BigData&AI Cell 인턴 (2026.07–): 주식등급평가모형 재구현 명세서 작성과 AI 보조 재구현 검증, 신용카드 스코어카드 모형 구현, 블록체인 데이터 기반 이상거래탐지(FDS) 분석
  • 연구 인천대학교 DIVE Lab 학부연구생 (2025.02–2026.02), 구매 전환 예측 논문 제1저자, Ubi-Media 2027 국제학술대회 논문 제2저자(발표 채택), 마케팅 피로도 인과추론 후속 연구 진행 중
  • 관심 Machine Learning · RecSys · GIS · Causal Inference · XAI

포트폴리오 · 이력서 · LinkedIn · Email

Python PyTorch scikit-learn Next.js

연구

동태적 마케팅 지식 기반 구매 전환 예측 · 제1저자, 『마케팅연구』(한국마케팅학회) 2026.11 게재 예정

전환율 0.1234%인 CRM 메시지 1억 7,200만 건에서 마케팅 이론 기반 피처 4묶음을 설계해 XGBoost PR-AUC를 0.3177에서 0.3620으로 높였습니다(9종 모형 중 최고). 상위 0.78% 메시지가 전환의 95.83%를 포착합니다(정밀도 15.26%, 무작위 대비 123.7배, holdout 기준).

대표 프로젝트

프로젝트 무엇을 했나 결과 (측정 조건)
DCN-V2와 행동 시퀀스 인코더를 활용한 CTR 예측 DCN-V2에 DIN·DIEN·BST 시퀀스 인코더를 붙여 비교 DIEN 검증 PR-AUC 0.0792(무작위 대비 약 4배), ROC-AUC 0.7413
신용카드 고객 세그먼트 분류 학습 240만 행·857개 컬럼의 결측 패턴별 처리와 스태킹 DACON private 0.6251(58위·상위 25%), 내부 표본 weighted F1 0.8936(다수 클래스 0.7026)

수상

대회 · 상격 · 날짜 프로젝트 내용
2024 임업통계 스마트활용 경진대회 · 대상(산림청장상) · 2024.12.13 산림 작물 적합지·생산량 분석 GIS 토양 레이어·기후·생산 데이터로 7개 작물의 적합지와 생산량 분석
2025 PBL 프로그램 · 대상(인천광역시장상) · 2026.02.27 동인천 원도심 상권 마케팅 파일럿 Marketing & Data Analysis 역할, 공실 데이터·상인 인터뷰 기반 30일 매장 파일럿
2025 인천대학교 산학 캡스톤디자인 경진대회 · 금상 · 2025.07.01 인천 e음 캐시백 정책 시뮬레이션 4인 팀 팀장, 예산 2,000억 원 안의 캐시백 일정 66개 비교

전체 11개 프로젝트는 포트폴리오에 있습니다.

포트폴리오 · 이력서 · LinkedIn · Email

English

I build prediction models and decision simulations from customer, policy, and spatial data, and state how far each result holds.

  • Now Intern, BigData&AI Cell, Hana Institute of Technology (Jul 2026–): reimplementation specification for an equity-rating model with AI-assisted reimplementation checks, a credit-card scorecard model, and fraud-detection (FDS) analysis on blockchain data
  • Research Undergraduate researcher, DIVE Lab, Incheon National University (Feb 2025–Feb 2026); first author of the conversion-prediction paper, second author of a paper accepted for presentation at Ubi-Media 2027, follow-up causal-inference study on marketing fatigue in progress
  • Interests Machine Learning · RecSys · GIS · Causal Inference · XAI

Portfolio · Resume · LinkedIn · Email

Research

Purchase Conversion Prediction with Dynamic Marketing Knowledge · first author, 마케팅연구 (Korean Marketing Association), forthcoming Nov 2026

On 172 million CRM messages with a 0.1234% conversion rate, four groups of theory-based features raised XGBoost PR-AUC from 0.3177 to 0.3620, the best of nine models. The top 0.78% of messages captured 95.83% of conversions (precision 15.26%, 123.7x lift over random, holdout).

Selected projects

Project What I did Result (setting)
CTR Prediction with DCN-V2 and Behavior-Sequence Encoders Compared DIN, DIEN, and BST sequence encoders on a DCN-V2 backbone DIEN validation PR-AUC 0.0792 (about 4x random), ROC-AUC 0.7413
Credit-Card Customer Segment Classification Missingness-aware handling and stacking over 2,400,000 training rows and 857 columns DACON private 0.6251 (58th, top 25%); internal-sample weighted F1 0.8936 vs 0.7026 majority baseline

Awards

Competition · Award · Date Project Work
2024 Forestry Statistics Smart Use Competition · Grand Prize (Korea Forest Service Commissioner Award) · 13 Dec 2024 Forestry Crop Suitability and Yield Analysis Suitability and yield analysis for seven crops from GIS soil layers, climate, and production data
2025 PBL Program · Grand Prize (Incheon Metropolitan City Mayor's Award) · 27 Feb 2026 Dong-Incheon Marketing Pilot Marketing & Data Analysis role; 30-day store pilot built on vacancy data and merchant interviews
2025 Incheon National University Industry-Academic Capstone Design Competition · Gold Prize · 1 Jul 2025 Incheon e-Eum Cashback Policy Simulation Team lead of four; compared 66 cashback schedules within a KRW 200 billion budget

All 11 projects are in the Portfolio.

Portfolio · Resume · LinkedIn · Email

Pinned Loading

  1. Resume Resume Public

    Resume of Junhyung Lee, data scientist (Korean / English)

  2. Portfolio Portfolio Public

    Data science & AI portfolio: 11 projects across CRM, RecSys, GIS, policy simulation, and LLM apps (EN/KO)

  3. forestry-crop-suitability forestry-crop-suitability Public

    Crop-site suitability for 7 forest crops: stat screening + per-crop SVM/RF/logit; historical AUC 0.84-0.94 (small tests)

    Jupyter Notebook

  4. incheon-eeum-policy-simulation incheon-eeum-policy-simulation Public

    Simulates 66 two-month cashback schedules on a KRW 200bn budget using district LightGBM forecasts; Apr+Jun ranked first

    Jupyter Notebook

  5. crm-conversion-prediction crm-conversion-prediction Public

    Purchase-conversion prediction with theory-based CRM features: XGBoost holdout PR-AUC 0.3620 at 0.12% conversion

    Jupyter Notebook

  6. DIVE-Research DIVE-Research Public

    Forked from thekimk/DIVE-Research

    HTML