This project is the final work in the "Machine Learning in Healtcare - 097248" course.
Ben Filiarsky & Eden Bar.
In this work, we propose a Hierarchical ML model for ICU mortality rate prediction. We believe that the ramification of certain health-conditions can aid the model’s ability to predict the mortality rate of ICU patients. Our approach is a novel take on an existing task - in-hospital mortality prediction, and one we strongly believe can alter the decision making process of ICU care and treatment trajectory. This is a proof of concept that building hierarchical models that include more conditions than just cardiac conditions might be beneficial for making better classification models.
Can be found in Project.pdf add link
- src - implementation of all neccesary modules for the project.
- hr_analysis.py - handle of signal loading, pre-processing and extraction of features from signals.
- load_data.py - sql handler for querying the MIMIC III dataset on an azure VM.
- load_sigs.py - utils file for hr_analysis.WaveForms class.
- prediction.py - module for training and evaluating different ML models.
- project.py - main file, sums up all the other files to a final implementation of the project's idea. Train and test ML models on predicting cardiac conditions within patients and later on predicting in-hospital mortality.
- queries.py - queries used to extract data from MIMIC III clinical dataset.
- data - relevant data for the training of the models and feature extraction.
- features_clinical.csv - clinical features for the patients in the sample.
- in_hospital_mortality.csv - table of dead patients and the admissions in which they died.
- matching.csv - matching table of patients with corresponding ECG signals entries.
- matching_numeric.csv - matching table of patients with corresponding numeric signal entries.
- wf_total.csv - features extracted from waveforms for all patients in the sample.