Semisupervised Classification With Novel Graph Construction for High-Dimensional Data.

Area of Science:

• Machine Learning
• Computer Vision
• Data Science

Background:

• Graph-based methods are effective for semi-supervised classification (SSC).
• Traditional methods suffer from predefined graphs and unsuitability for high-dimensional, noisy data.
• Existing approaches fail to integrate classifier training with similarity matrix learning.

Purpose of the Study:

• To propose a novel SSC method (SSC-NGC) for robust and accurate classification.
• To address limitations of traditional graph-based SSC methods.
• To develop a unified framework for adaptive learning of classifier, graph, and subspace.

Main Methods:

• Proposed SSC-NGC method optimizes similarity matrices in label space and an additional subspace.
• Learned projection matrix preserves local and global data structure for high-quality subspace.
• Integrated classifier training, graph construction, and subspace learning into a unified framework.

Main Results:

• The proposed method achieved superior performance compared to state-of-the-art algorithms.
• Experimental results on multiple real-world datasets validated the method's effectiveness.
• The unified framework adaptively learned optimal joint results through an iterative scheme.

Conclusions:

• SSC-NGC offers a more robust and accurate approach to semi-supervised classification.
• The method excels in handling high-dimensional data with noisy or redundant features.
• The integrated framework provides adaptive learning for improved classification outcomes.