CARE: Finding Local Linear Correlations in High Dimensional Data.

Area of Science:

• Data Science
• Machine Learning
• Bioinformatics

Background:

• High-dimensional data analysis is crucial for many applications, including biomedicine.
• Existing methods often focus on global patterns in the full feature space.
• Local latent patterns within feature subsets are often missed by global approaches.

Purpose of the Study:

• To investigate the problem of finding local linear correlations in high-dimensional data.
• To identify latent pattern structures existing in specific subspaces.
• To develop an algorithm capable of finding strongly correlated feature subsets supported by a significant portion of data points.

Main Methods:

• The study formalizes the problem of finding local linear correlations.
• An algorithm named CARE (Correlation Aware REgression) is proposed.
• CARE utilizes spectral properties and heuristic approaches to efficiently prune the search space.

Main Results:

• Extensive experiments demonstrate the effectiveness of the CARE algorithm.
• CARE successfully identifies local linear correlations missed by existing methods.
• The algorithm proves adept at uncovering subspace-specific patterns.

Conclusions:

• The proposed CARE algorithm offers a powerful new tool for high-dimensional data analysis.
• It enables the discovery of local latent patterns vital for biomedical research.
• CARE provides a more comprehensive approach to understanding complex datasets.