Utility metric for unsupervised feature selection.

Area of Science:

• Data Science
• Machine Learning
• Computational Statistics

Background:

• Feature selection is crucial for dimensionality reduction and data interpretability.
• Unsupervised feature selectors are necessary for data lacking annotations.
• Existing unsupervised methods are often complex, computationally expensive, and require parameter tuning.

Purpose of the Study:

• To propose a publicly available, ready-to-use unsupervised feature selector.
• To achieve comparable results to state-of-the-art methods with lower computational cost.
• To eliminate the need for parameter tuning in unsupervised feature selection.

Main Methods:

• The study proposes the U2FS algorithm, a spectral feature selection method.
• It involves manifold learning using a radial basis function (RBF) kernel with an alternative parameter estimation for high-dimensional data.
• Subset selection is performed using a backwards greedy approach with a least-squares utility metric.

Main Results:

• U2FS successfully selects relevant features in simulation environments.
• Performance on benchmark datasets is comparable to existing state-of-the-art methods.
• U2FS demonstrates significantly lower computational time compared to other methods.

Conclusions:

• U2FS provides an effective and efficient unsupervised feature selection solution.
• The algorithm is accessible due to its ready-to-use nature and lack of parameter tuning.
• U2FS advances spectral feature selection techniques for high-dimensional data analysis.