LSTM based early diagnosis of Sepsis-3

Objective

The gaol of final project for CSE 6240 Big Data for Health was to use Big Data Tools for solving a HealthCare problem. The problem definition was to develop a machine-learning based predicitve model for early detection of Sepsis which is a medical condition where the immune system damages the body as a result of fighting infection.

Methodology:

• Use SQL and PySpark to caluclate features from the Medical Information Mart for Intensive Care (MIMIC)-III dataset
• Replicate and train a Long Short-Term Memory (LSTM) neural network model using the calculated features
• Experimental evaluation and parameter tuning
• Present findings in a report

Code

github repo

Tools & Technologies

MIMIC-III Clinical Database, Google Big Query, SQL, PySpark, Feature Engineering, PyTorch, LSTM