This repository provides a simplified, modular, and reproducible pseudocode implementation of the machine learning framework used in the study:
"Machine learning-based utilization of lipid panels improves the predictability of kidney dysfunction"
Soie Kwon*, Donghwan Yoon*, Changhwa Park*, Sehoon Park, Yong Chul Kim, Dong Ki Kim, Kook-Hwan Oh, Kwon Wook Joo, Yon Su Kim, Sungroh Yoon†, Seung Seok Han†
*Co-first authors, †Co-corresponding authors
This repository was developed to demonstrate the core logic of the machine learning methodology presented in the paper using a reproducible pseudocode format. It enables simplified benchmarking of models trained on simulated lipid panel data for predicting kidney outcomes.
simulate_data.py: Simulates static and time-series patient data with lipid variables and outcome labels.model.py: Defines MLP, RNN, and machine learning models (Logistic Regression, Random Forest, LightGBM).utils.py: Contains helper functions for evaluation, dataloading, and data splitting.main.py: Runs the entire pipeline (data generation → model training → evaluation).
This code is intended solely for methodological demonstration and does not include real clinical data.