Ensemble-Enhanced Semi-Supervised Learning With Optimized Graph Construction for High-Dimensional Data.

Area of Science:

• Machine Learning
• Data Science
• Computer Vision

Background:

• Graph-based methods excel in semi-supervised classification but struggle with high-dimensional data due to limitations in utilizing prior information and capturing intrinsic data distribution.
• Existing approaches often construct graphs in either the original or output space, failing to fully leverage data characteristics.

Purpose of the Study:

• To introduce a novel approach, Semi-Supervised Classification with Optimized Graph Construction (SSC-OGC), for improved semi-supervised classification.
• To develop a Hybrid Subspace Ensemble-enhanced framework (HSE-OGC) that overcomes the limitations of traditional graph construction in high-dimensional spaces.

Main Methods:

• SSC-OGC utilizes both predefined and adaptive graphs, incorporating graph constraint regularization (GCR) and collaborative constraint regularization (CCR) to enhance graph structure and subspace learning.
• HSE-OGC constructs multiple hybrid subspaces with selected features, creating diverse representations. Multiple predefined graphs are built within these subspaces, and SSC-OGC classifiers are trained in an ensemble manner.

Main Results:

• Experimental results on various high-dimensional datasets demonstrate the superior performance of the proposed HSE-OGC method.
• The integration of hybrid subspaces and ensemble learning significantly boosts overall classification accuracy.

Conclusions:

• HSE-OGC offers a robust solution for semi-supervised classification in high-dimensional, noisy datasets.
• The proposed method effectively leverages prior information and captures optimal intrinsic data distributions through optimized graph construction and ensemble strategies.