/Leonhard

The Predictive Power of Biosignals for Cardiac Arrhythmias

Primary LanguagePythonMIT LicenseMIT

In this work, we strive to find a quantitative basis for more reliable diagnoses by attempting to measure the importance of specific factors potentially relevant to the diagnosis of cardiac arrhythmias. Using Random Forest in conjunction with data from the MIMIC-III database, we made predictions about patient diagnosis and measured feature importance. We collected data on intensive care unit (ICU) patients over the age of 16 and tested the importance of respiratory rate, blood pressure, sodium, potassium, calcium, among other features. The influence of the quantity of data was also measured by adjusting the amount of time over which data was collected. The model achieved, at its best, an Area Under the Receiver Operator Curve (AUC) score of 0.9787 and, thus, confirmed the importance of several previously suggested factors in the diagnosis of cardiac arrhythmias.  The files associated wth this project can be used to recreate our experiments with the use of NumPy, Dateutil, Sci-Kit Learn, and ArgParse packages.