Sepsis_Detection

Predictive Septic Shock for ICU patients

This was a project I did for my Insight September 2020 session. The purpose of this project is to create an algorithm that will predict whether or not a patient will go into septic shock prematurely! The idea is to identify as early as possible if an ICU patient will go into septic shock.

Data

The date comes from the MIMIC IV repository

• https://mimic-iv.mit.edu/

For this project approximately 10 million rows were obtained from MIMIC IV. This was narrowed down to 2 million total entries. With approximately 40,000 patients. Root files in this directoy include data aggregation and EDA

Folder Files

• v1 lstm with 3 units for a time i window of 100 - model just predicts the moajority class\
• v2 same lstm, but with the correct labels, still the same as v1
• v3 experimenting with xgboost to identify feature importance values 25Sep20
• v4 experimenting with Cox Models and Surivial Analysis 28Sep20
• v5 experimenting with LSTM-Attention base models 01Sep20