A weight clustering algorithm based on sliding window model for stream data.

Area of Science:

• Data Science
• Machine Learning
• Algorithm Development

Background:

• Streaming data presents significant challenges for traditional clustering due to temporal variations and large volumes.
• Concept drift, a common phenomenon in streaming data, further complicates accurate cluster analysis.
• Existing clustering methods often struggle to adapt to the dynamic nature of data streams.

Purpose of the Study:

• To propose a novel weighted clustering approach tailored for streaming data.
• To address the challenges posed by concept drift in data streams.
• To enhance clustering accuracy and efficiency in dynamic environments.

Main Methods:

• In-depth analysis of concept drift features in streaming data.
• Development of a weight parameter calculation technique.
• Implementation of a sliding window model clustering algorithm with threshold calculations.
• Two-stage clustering: intra-window clustering and landmark window cluster merging.

Main Results:

• On static datasets, the algorithm demonstrated low runtime and misclassification rates but struggled with precise clustering.
• On concept-drifting datasets, the algorithm achieved accurate clustering with minimal misclassification rates when appropriate weight parameters were used.
• The sliding window approach effectively adapts to dynamic data characteristics.

Conclusions:

• The proposed weighted clustering approach is effective for streaming data analysis, particularly in the presence of concept drift.
• The algorithm shows adaptability to dynamic data environments, offering potential for real-world applications.
• Optimizing weight parameters is crucial for achieving high accuracy in concept-drifting scenarios.