Locality preserving score for joint feature weights learning.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Locality preserving criteria are common for feature selection quality assessment.
• Existing unsupervised methods struggle with similarity matrix consistency and feature redundancy.

Purpose of the Study:

• To propose a novel unsupervised feature selection algorithm addressing current limitations.
• To jointly learn adaptive nearest neighbors in the weighted space for improved feature selection.

Main Methods:

• Developed a novel algorithm for unsupervised feature selection.
• Employed an iterative approach solving convex subproblems efficiently.
• Learned adaptive nearest neighbors within the weighted feature space.

Main Results:

• The proposed algorithm demonstrated superior performance compared to state-of-the-art methods.
• Experiments on UCI and face datasets confirmed its effectiveness.
• Achieved higher classification accuracy than existing unsupervised and supervised techniques.

Conclusions:

• The novel approach effectively overcomes weaknesses in traditional locality preserving feature selection.
• Jointly learning adaptive neighbors in weighted space enhances feature selection quality.
• The method offers a robust solution for improving classification accuracy.