Consensus Clustering With Co-Association Matrix Optimization.

Area of Science:

• Computational Biology
• Data Science
• Machine Learning

Background:

• Consensus clustering integrates multiple data partitions for robust results.
• Existing methods often lack optimization, ignore original data, and overlook noise.

Purpose of the Study:

• To propose a novel consensus clustering method (CC-CMO) for improved co-association matrix optimization.
• To address limitations of existing methods by incorporating label and feature space information and noise reduction.

Main Methods:

• CC-CMO utilizes a weighted partition matrix and least squares regression (LSR) in label space.
• In feature space, it minimizes reconstruction error via doubly stochastic normalization for noise elimination and local affinity learning.
• A unified optimization framework with an alternating optimization algorithm is employed.

Main Results:

• CC-CMO effectively integrates subspace representation, global structure, and local affinity.
• The method demonstrates superior performance compared to state-of-the-art consensus clustering approaches.
• Extensive experiments on diverse real-world datasets validate the proposed method's effectiveness.

Conclusions:

• CC-CMO offers a robust and optimized approach to consensus clustering.
• The method enhances clustering accuracy by strategically leveraging data from both label and feature spaces.
• CC-CMO represents a significant advancement in consensus clustering techniques.