Big Data Clustering via Community Detection and Hyperbolic Network Embedding in IoT Applications.

Area of Science:

• Computer Science
• Data Science
• Network Science

Background:

• Big sensory data from Internet of Things (IoT) applications presents significant clustering challenges.
• Existing data clustering methods struggle with the scale and complexity of multi-dimensional IoT datasets.
• Mapping data clustering to community detection on data graphs offers a promising approach.

Purpose of the Study:

• To propose a novel data clustering framework for big sensory data from IoT applications.
• To enhance the Girvan-Newman (GN) community detection algorithm for efficient node clustering on large data graphs.
• To demonstrate the framework's efficacy on artificial networks and real-world IoT data.

Main Methods:

• Representing multi-dimensional IoT data as a network graph.
• Mapping data clustering to a community detection problem on the data graph.
• Enhancing the Girvan-Newman algorithm using hyperbolic network embedding (Rigel) for efficient edge-betweenness centrality computation.
• Evaluating the approach on scale-free, small-world, and random geometric network topologies and benchmark datasets.
• Applying the framework to multi-dimensional datasets from a smart-city/building IoT infrastructure (FIESTA-IoT).

Main Results:

• The proposed hyperbolic network embedding significantly enhances the efficiency of the Girvan-Newman algorithm for node clustering.
• The framework achieves efficient data clustering in terms of modularity without substantial accuracy loss.
• Proof-of-concept evaluation on FIESTA-IoT data confirms the framework's applicability to real-world IoT scenarios.
• The approach demonstrates effectiveness across various artificial network types.

Conclusions:

• The novel framework provides an efficient and accurate method for data clustering of big sensory data in IoT.
• Hyperbolic network embedding offers a scalable solution for community detection on large data graphs.
• The framework has potential applications in energy-efficient smart-city/building sensing and other IoT domains.