A density-based competitive data stream clustering network with self-adaptive distance metric.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Data Mining

Background:

• Data stream clustering involves analyzing ordered patterns in sequential data.
• Existing methods may struggle with arbitrarily shaped clusters and noisy data.
• Unsupervised learning is crucial for discovering patterns without predefined labels.

Purpose of the Study:

• To propose an unsupervised learning neural network for effective data stream clustering.
• To address limitations of current methods in handling complex cluster shapes and noise.
• To develop a model capable of learning from unnormalized input data.

Main Methods:

• Introduction of the Density Based Self Organizing Incremental Neural Network (DenSOINN).
• DenSOINN employs a self-organizing, incrementally growing competitive network architecture.
• Utilizes a competitive Hebbian learning rule for online unsupervised and topology learning.
• Incorporates a density-based clustering mechanism and a self-adaptive distance framework.

Main Results:

• DenSOINN successfully discovers arbitrarily shaped clusters in data streams.
• The model effectively mitigates the negative impact of noise on clustering accuracy.
• Achieves high performance in learning from unnormalized input data.
• Experimental results demonstrate superior performance compared to state-of-the-art methods.

Conclusions:

• DenSOINN offers a robust solution for data stream clustering tasks.
• The proposed network architecture and mechanisms enhance cluster discovery and noise resilience.
• DenSOINN shows significant potential for real-world applications involving sequential data analysis.