Unsupervised Feature Selection via Orthogonal Basis Clustering and Local Structure Preserving

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• High-dimensional data presents challenges due to the "curse of dimensionality."
• Feature selection is crucial for reducing redundancy and identifying informative features.
• Unsupervised feature selection is vital for preprocessing data without prior category information, applicable in pattern classification and learning systems.

Purpose of the Study:

• To propose an efficient unsupervised feature selection method named OCLSP.
• To address the challenge of selecting informative features in high-dimensional datasets.
• To simultaneously achieve excellent cluster separation and preserve local data structures.

Main Methods:

• Developed OCLSP using orthogonal basis clustering and adaptive graph regularization.
• Employed an efficient alternative optimization algorithm to solve the method's optimization problem.
• Conducted theoretical analysis of computational complexity and convergence.

Main Results:

• OCLSP demonstrated superior performance on nine real-world datasets.
• Achieved higher clustering accuracy and normalized mutual information compared to state-of-the-art methods.
• Showcased a strong ability to identify the most important features in datasets.

Conclusions:

• The proposed OCLSP method is effective for unsupervised feature selection in high-dimensional data.
• OCLSP offers significant improvements over existing unsupervised feature selection techniques.
• The method shows promise for applications requiring robust feature identification and data preprocessing.