AW-GBGAE: An Adaptive Weighted Graph Autoencoder Based on Granular-Balls for General Data Clustering

Area of Science:

• Data Science
• Machine Learning
• Artificial Intelligence

Background:

• High-dimensional, unlabeled data often contains intrinsic relationships suitable for graph-based clustering.
• Challenges include a lack of edge structure information and the presence of irrelevant features.

Purpose of the Study:

• To develop a robust graph-based clustering method for high-dimensional unlabeled data.
• To address limitations of existing methods in handling feature relevance and edge construction.

Main Methods:

• A feature weighting approach is applied to manage features.
• Edges are constructed using weighted granular-balls.
• Graph convolutional networks (GCNs) are integrated with edge generation within an autoencoder network (AW-GBGAE).

Main Results:

• The proposed AW-GBGAE method significantly enhances information extraction from high-dimensional, unlabeled data.
• Experimental results show superior performance in clustering tasks compared to baseline models.
• The model demonstrates strong competitiveness and reliability.

Conclusions:

• AW-GBGAE offers an effective solution for clustering complex, high-dimensional datasets.
• The integration of feature weighting, granular-balls, and GCNs improves clustering performance.
• The method provides a reliable and competitive approach for information extraction.