Hybrid Sampling-Based Clustering Ensemble With Global and Local Constitutions.

Area of Science:

• Machine Learning
• Data Mining
• Pattern Recognition

Background:

• Boosting and bagging are popular ensemble methods for classification.
• Boosting excels on noise-free data with complex structures, while bagging is robust to noisy data.
• Existing ensemble methods are primarily focused on classification, with limited extensions to clustering.

Purpose of the Study:

• To extend boosting and bagging ensemble techniques to clustering tasks.
• To propose a novel hybrid sampling-based clustering ensemble method.
• To develop a consensus function that integrates local and global cluster structures.

Main Methods:

• Iterative generation of input partitions using a hybrid boosting-bagging process.
• Development of a novel consensus function to represent cluster structures.
• Application of a single clustering algorithm to the consensus representation for a final partition.

Main Results:

• The proposed hybrid ensemble clustering method was evaluated on synthetic, benchmark, and facial recognition datasets.
• The technique demonstrated superior performance compared to existing benchmarks.
• The method effectively handles both local and global cluster structures.

Conclusions:

• The novel hybrid ensemble clustering approach successfully combines the strengths of boosting and bagging.
• The proposed consensus function is effective in consolidating diverse cluster structures.
• This method offers improved performance for various clustering tasks, including facial recognition.