A new Growing Neural Gas for clustering data streams

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Clustering massive datasets presents challenges in memory and time constraints.
• Efficiently partitioning continuous data streams is crucial for big data analysis.

Purpose of the Study:

• To introduce G-Stream, a novel algorithm for clustering data streams.
• To enable efficient, single-pass clustering of massive datasets without prior cluster number assumptions.

Main Methods:

• The G-Stream algorithm utilizes a growing neural gas approach.
• It incorporates a reservoir and a fading function to enhance clustering quality.
• The algorithm processes data streams in a single pass.

Main Results:

• G-Stream effectively clusters data streams with arbitrary shapes.
• The use of a reservoir and fading function improves clustering performance.
• The algorithm demonstrates efficiency on public datasets.

Conclusions:

• G-Stream offers an efficient solution for clustering data streams.
• The algorithm's adaptability to arbitrary cluster shapes and its performance make it suitable for massive datasets.