Community detection: Comparison among clustering algorithms and application to EEG-based brain networks.

Area of Science:

• Network Science
• Computational Neuroscience
• Data Analysis

Background:

• Community structure is vital for understanding complex network organization, especially in brain networks.
• Numerous clustering algorithms exist, but consensus on the most reliable is lacking.
• Comparative analysis is needed to guide algorithm selection for network research.

Purpose of the Study:

• To comparatively analyze the performance of six community detection algorithms.
• To evaluate algorithm reliability across diverse network conditions (noise, density, size).
• To provide guidelines for selecting appropriate algorithms based on network characteristics.

Main Methods:

• Simulated networks with varying properties were used as ground truth.
• Six distinct clustering algorithms were tested on these simulated networks.
• Algorithm performance was assessed based on factors like noise level and network density.

Main Results:

• Algorithm performance varied significantly with network properties like noise and density.
• Louvain and Leicht & Newman algorithms demonstrated superior performance across most conditions.
• Ronhovde and Infomap algorithms were more effective in highly noisy network environments.

Conclusions:

• Algorithm choice significantly impacts community detection in complex networks.
• Louvain and Leicht & Newman are recommended for general use, while Ronhovde and Infomap suit noisy data.
• The findings were validated by applying algorithms to EEG-derived brain functional connectivity networks.