Comparative Analysis of Feature Selection Methods in the use-case of ARDS Classification in Clinical Time-Series Data

Abstract

This study investigates four feature selection methods (?2, ANOVA F-test, Lasso, and a Tree-based method) to enhance Acute Respiratory Distress Syndrome (ARDS) classification in time-series intensive care unit data using an existing Random Forest algorithm. While feature selection did not significantly improve ARDS classification performance, using a reduced number of features achieved comparable results to using the entire dataset. This indicates the potential of dimensionality reduction for ARDS classification.