Self-Weighted Clustering With Adaptive Neighbors.

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• Modern clustering often involves separate steps for similarity graph construction and partitioning.
• Existing methods frequently assume equal feature importance, limiting practical applicability.
• Learning effective similarity graphs is crucial for robust clustering.

Purpose of the Study:

• To propose a novel clustering model that simultaneously learns feature weights and similarity graphs.
• To address the limitation of equal feature importance in traditional clustering approaches.
• To enhance clustering performance by adaptively weighting features.

Main Methods:

• Developed a self-weighted clustering with adaptive neighbors (SWCAN) model.
• Integrated feature weighting, similarity graph learning, and sample partitioning into a unified framework.
• Evaluated the model's ability to assign appropriate weights to different features.

Main Results:

• The SWCAN model demonstrated the capability to assign meaningful weights to features.
• SWCAN significantly outperformed existing clustering models on both synthetic and real-world datasets.
• Experimental results validate the effectiveness of adaptive feature weighting in clustering.

Conclusions:

• The proposed SWCAN model offers a more robust and adaptable approach to clustering.
• Simultaneous learning of feature weights and similarity graphs improves clustering accuracy.
• SWCAN provides a valuable advancement for practical clustering applications where feature importance varies.