Publications

2026

1. Graph Attention Neural Networks for Survival Analysis of Default in Peer-to-Peer Lending

   Min-seok Kim and Jooyoung Lee

   Chung-Ang University, 2026

   Abs Bib

   This study proposes CoxGAT, a graph attention network-based survival model for predicting time-to-default in P2P lending. By constructing a mutual KNN borrower graph and integrating attention-based message passing with Cox partial likelihood optimization, CoxGAT achieves a C-index of 0.705, outperforming CoxPH (0.550), RSF (0.688), and CoxMLP (0.693). The results demonstrate that financial risk is shaped not only by individual borrower attributes but also by relational structures among borrowers with similar characteristics.

   Liquid error: internal

2. Oncological and Perioperative Outcomes of Robot-Assisted Radical Cystectomy: A Real-World Cohort Study Using Cloning Approach

   Min-seok Kim, Jooyoung Lee, yongseong Lee, Tuan Thanh Nguyen, and Seyoung Choi

   2026

   In preparation

   Abs Bib

   This study evaluates oncological and perioperative outcomes of robot-assisted radical cystectomy compared to open surgery using a cloning approach in a real-world cohort.

   Liquid error: internal

3. Machine learning for gait data improved classification of balance status using multivariate functional PCA combined with square root velocity framework

   Min-seok Kim, Jooyoung Lee, Jieun Cho, and Hogene Kim

   2026

   In preparation

   Abs Bib

   This study analyzed walking and balance characteristics of stroke patients using Multivariate functional principal component analysis and SRVF to identify which motion patterns affect storke patients.

   Liquid error: internal

2025

1. Standing balance test predicts the Berg Balance Scale score in patients with stroke using principal component analysis

   Jieun Cho, sunghe Ha, Jooyoung Lee, Min-seok Kim, and Hogene Kim

   Scientific Reports, 2025

   Abs DOI Bib HTML

   This study investigated the relationship between standing balance test results and Berg Balance Scale scores in stroke patients using principal component analysis.

   Liquid error: internal

2. Comparison of Overall Survival After Open vs. Robot-Assisted Radical Cystectomy: A Target Trial Emulation Approach to Mitigate Immortal Time Bias

   Min-seok Kim and Jooyoung Lee

   The Korean Journal of Applied Statistics, 2025

   Abs Bib

   This study compared overall survival outcomes between open and robot-assisted radical cystectomy using a target trial emulation approach to address immortal time bias in observational data.

   Liquid error: internal

2024

1. Stroke walking and balance characteristics via principal component analysis

   Jieun Cho, sunghe Ha, Jooyoung Lee, Min-seok Kim, and Hogene Kim

   Scientific Reports, 2024

   Abs DOI Bib HTML

   This study analyzed walking and balance characteristics of stroke patients using principal component analysis to identify key movement patterns.

   Liquid error: internal

2. Modeling for Flood Prediction: Diverse Applications and Evaluation of Machine Learning and Deep Learning Approaches

   Min-seok Kim, Suyoung Park, Yunseo Lee, and Ri Han

   데이터과학연구, 2024

   Abs Bib

   This study compares regression, machine learning (LGBM, CatBoost, XGBoost, AutoML), and deep learning (TabNet, TabPFN, SAINT) approaches for flood probability prediction. The effect of PCA and StandardScaler preprocessing was evaluated across all models. A weighted ensemble of machine learning models (LGBM, CatBoost, XGBoost at 8:2 ratio) achieved the highest performance with R² = 0.867.

   Liquid error: internal