ICU Intubation Prediction
GitHub Link
Tags
Machine Learning
Feature Selection
EDA
Date
Summary
Summary: Based on the exploratory data analysis on the pre-extracted MIMIC-IV dataset, applied SOMTE as the oversampling method to deal with the imbalanced data. Then implemented Logistic Regression, Decision Tree and Random Forest classifier for the prediction. For each model, applied Genetic Algorithm for feature selection separately and used grid search to finetune the parameters of each model. The Random Forest classifier achieve the highest Area Under ROC of 0.76 with an accuracy of 0.96 on predicting the patient’s chance of intubation in ICU.