Reproduce “Feature Selection: A Data Perspective” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Jundong Li, Kewei Cheng, Suhang Wang, Fred Morstatter, Robert P. Trevino, Jiliang Tang, Huan Liu

This survey paper provides a comprehensive, data-centric re-evaluation of feature selection techniques, proposing a novel taxonomy that categorizes algorithms based on the characteristics of conventional, structured, heterogeneous, and streaming data, alongside releasing the `scikit-feature` open-source repository.

Key insights

  • Feature selection techniques can be effectively categorized and understood through a "data perspective," unifying traditional methods with modern approaches for complex data types.
  • Many diverse feature selection algorithms can be unified under broader theoretical frameworks, such as utility maximization for similarity-based methods and conditional likelihood for information theoretical methods.
  • Effectively addressing high-dimensional data requires specialized feature selection methods tailored for structured, heterogeneous, and streaming data, moving beyond conventional flat-feature approaches.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.