Code for the KDD'26 paper "ClinicalBench: Can LLMs Beat Traditional ML Models in Clinical Prediction?"

Advanced machine learning system for cardiovascular disease risk prediction using clinical biomarkers and patient health indicators

Published in Artificial Intelligence in Health (2026) AIH. Leakage-resistant ML for cardiovascular risk prediction in diabetic patients. DOI: 10.36922/AIH025490111

COVERT: private, communication-efficient vertical federated block-term tensor regression on clinical multi-view data (imaging, ECG, labs), at parity with centralized while transmitting only a DP-noised coupling score.

Machine learning models for predicting type 1 diabetes complications — retinopathy, nephropathy, neuropathy, diabetic foot (Python, scikit-learn, XGBoost)

Competing-risks survival re-examination of the Mayo PBC cirrhosis benchmark (UCI Cirrhosis dataset): a leakage-free, reproducible pipeline benchmarking seven models against the Mayo clinical score.

Open-source, reproducible LSTM implementation for in-hospital mortality prediction with Focal Loss and calibration

ICU LOS prediction for pneumonia patients using MIMIC-III time-series data & interpretable ML

Antenatal preterm-birth risk model on CDC NVSS Natality data — TRIPOD+AI reporting, temporal external validation, leakage guard, calibration, decision-curve analysis, SHAP, and subgroup fairness. Fully reproducible.

Clinical predictive model for fetal health classification from cardiotocogram data using logistic regression in SAS. Supports early intervention and maternal-fetal health monitoring.

GCE framework for interpretable modeling in cardiac sarcoma survival using machine and deep learning. It captures complex feature relationships to enhance predictive insights and clinical understanding.

Predicting early functional mobility decline using longitudinal biomarker trajectories from OMOP EHR data (AMIA submission).

Evaluation of missing data imputation strategies for cardiovascular risk prediction using MIMIC-IV electronic health records and the AHA PREVENT equation.

WiDS Datathon 2026 clinical prediction pipeline with gradient boosting ensembles, calibration, and leaderboard score 0.97089

Foundation model for health trajectory prediction — architecture shootout across Transformer, Mamba, and Continuous-Time

Interpretable statistical learning for 10-year coronary heart disease risk prediction using logistic regression, cross-validated LASSO, calibration analysis, and bootstrap uncertainty estimation.

End-to-end ML project: 30-day hospital readmission prediction using XGBoost, PostgreSQL, Python and Tableau

Reference implementation of MicrobiomeRiskScore (MRS): a treatment-aware Transformer that predicts healthcare-associated infections from longitudinal gut microbiome dynamics. Companion code to Ma et al., npj Digital Medicine.

Longitudinal validation of metabolic syndrome risk scores on NHANES Linked Mortality File

TrajGPT vs CLMBR-T-base on EHRSHOT: training, embedding extraction, and few-shot evaluation across 15 clinical prediction tasks.