Minimum Redundancy Maximum Relevance (mRMR) is a feature selection algorithm that identifies the most informative subset of features by maximizing their relevance to the target variable while minimizing redundancy among the selected features themselves. In WiFi/CSI sensing research, it is used to distill high-dimensional CSI measurements — such as subcarrier amplitude and phase values across multiple antennas — into compact, discriminative feature sets, improving model accuracy and reducing computational overhead for tasks like device-free crowd counting. The method is typically applied as a preprocessing step before classification or regression, and may be used alongside or compared to other selection techniques such as Principal Component Analysis (PCA) or filter-based variance thresholding.

Source Papers

  • CSI crowd-counting: An experimental study using Machine Learning and Deep Learning Algorithms — CSI crowd-counting: An experimental study using Machine Lear
  • FreeCount: Device-Free Crowd Counting with Commodity WiFi — FreeCount: Device-Free Crowd Counting with Commodity WiFi