The least sample size essential for detecting changes in clustering solutions of streaming datasets

Area of Science:

• Data Science
• Machine Learning
• Statistics

Background:

• Clustering analysis groups data based on similarity, but evolving data streams present challenges.
• Existing research monitors changes in cluster solutions for dynamic data, yet overlooks cluster size variability.
• No guidelines exist on how cluster size impacts changes observed in evolving data streams.

Purpose of the Study:

• To examine the evolution of clusters in dynamic scenarios concerning variability in cluster sizes.
• To address the gap in understanding the effect of cluster size variability on evolving cluster solutions.
• To determine the minimum sample size for effective clustering of time-stamped datasets.

Main Methods:

• A simulation study using artificial datasets.
• Analysis of cluster evolution in response to varying cluster sizes.
• Investigation into the relationship between cluster size and observed changes in data streams.

Main Results:

• Variability in cluster sizes significantly impacts the evolution of cluster solutions in dynamic data streams.
• Specific changes in cluster solutions are demonstrably influenced by the sizes of the clusters involved.
• The study identifies minimum sample size requirements for reliable clustering of time-stamped data.

Conclusions:

• Understanding cluster size variability is essential for accurately monitoring evolving data.
• The findings provide crucial insights for developing robust algorithms for dynamic clustering.
• This research offers practical guidance on sample size determination for time-series data analysis.