An incremental clustering method based on the boundary profile.

Area of Science:

• Data Mining and Machine Learning
• Big Data Analytics
• Clustering Algorithms

Background:

• Continuous data generation in applications like finance and satellite monitoring necessitates efficient Big Data clustering.
• Reclustering large datasets from scratch is computationally expensive and time-consuming.
• Incremental clustering methods are crucial for adapting to evolving datasets.

Purpose of the Study:

• To propose a novel incremental clustering method, Boundary-Profile-Based Incremental Clustering (BPIC), for dynamically growing datasets.
• To develop a Boundary-Vector-Based Boundary Point Detection (BV-BPD) algorithm for summarizing cluster structures.
• To enhance clustering efficiency and accuracy for large, evolving datasets.

Main Methods:

• The BPIC method represents clusters using boundary profiles, reducing storage and processing costs.
• A BV-BPD algorithm identifies boundary points to summarize cluster structures.
• New data points are processed online, with updates occurring in batches, using a bucket system to manage noise and potential new clusters.

Main Results:

• The BV-BPD algorithm demonstrated superior performance compared to state-of-the-art methods in boundary point detection.
• The BPIC method achieved qualified clustering results on large datasets.
• BPIC exhibited higher time and space efficiency compared to other incremental clustering methods.

Conclusions:

• The BPIC method offers an efficient and robust solution for incremental clustering of Big Data.
• Representing clusters by boundary profiles significantly reduces computational and storage overhead.
• The method's insensitivity to noise and data order ensures reliable clustering performance.