Nuclear Norm Clustering: a promising alternative method for clustering tasks.

Area of Science:

• Data Science
• Machine Learning
• Bioinformatics

Background:

• Clustering is essential for identifying patterns in data, but many methods struggle with noise and outliers.
• Existing clustering techniques like k-means have limitations in robustness and sensitivity to data imperfections.
• There is a need for advanced clustering algorithms that can handle real-world, noisy datasets effectively.

Purpose of the Study:

• To introduce Nuclear Norm Clustering (NNC) as a robust and effective clustering algorithm.
• To present NNC as a competitive alternative to traditional methods such as k-means.
• To evaluate NNC's performance on diverse datasets, including genomic data.

Main Methods:

• Nuclear Norm Clustering (NNC) algorithm utilizing a data matrix (M) and a desired number of clusters (K).
• Simulated annealing techniques employed to optimize a label vector, minimizing the nuclear norm of the pooled within-cluster residual matrix.
• Comparative analysis against classic clustering methods using 15 public benchmark datasets and 2 psoriasis genome-wide association studies (GWAS).

Main Results:

• NNC demonstrated competitive performance, achieving high F-scores across 15 benchmarked public datasets.
• The NNC algorithm showed strong results on 2 psoriasis genome-wide association studies (GWAS) datasets.
• NNC proved effective in handling noisy data and outliers, outperforming traditional clustering approaches.

Conclusions:

• Nuclear Norm Clustering (NNC) is a promising and robust alternative for various clustering applications.
• NNC offers improved performance, particularly in scenarios with noisy or outlier-prone data.
• The algorithm's effectiveness is validated by its performance on both general and specialized genomic datasets.