Problem description
Prediction of patient mortality from just the number of patient-hospital interactions per day.
Problem source (URL)
https://www.kaggle.com/datasets/drscarlat/mimic3c
Codebase description
This codebase first spot-checks multiple machine learning algorithms to decide which one is most accurate for this problem. After selecting the random forest classifier as the best method, it then sets the model parameters and trains the model.
This codebase also utilises NN (neural network) for the same task so it is possible to compare two different approaches to the same problem.
Codebase source (URL)
https://www.kaggle.com/code/drscarlat/predict-hospital-mortality-mimic3/notebook
Dataset description
This dataset is an ICU data aggregated as the number of interactions between patient and hospital. The original data is from MIMIC3 - Multiparameter Intelligent Monitoring in Intensive Care (de identified DB) available freely from https://physionet.org/
Each row in the csv file is one admission. The independent variables are age, gender, admission type, admission source, daily average number of: labs, micro labs, IV meds, non-IV meds, imaging reports, notes, orders, caregivers, care units, etc. and the label is hospital mortality. The LOS was removed from the features as it may give a hint on the patient outcome.
Dataset source (URL)
