KaiWU17TUM/Risk-Estimation-for-ICU-Patients-with-Personalized-Anomaly-Encoded-Patient-Bedside-data

EMBC2023

Risk-Estimation-for-ICU-Patients-with-Personalized-Anomaly-Encoded-Patient-Bedside-data

Data preparation

Our experiements are conducted with MIMIC-IV 1.0 database. To run the code, you need to

• download the database;
• follow this tutorial to load MIMIC-IV into PostgreSQL database;
• run notebooks/generate_all_mimic_dataset.ipynb to read and save relevant data from the database;
• run src/data_prep.py to generate temporally-aligned 48H data samples in raw and anomaly-encoded format.

Train models

• Run train_cnn.sh to train CNN models
• Run train_xgb.sh to train RandomForest or XGBoost models