Cluster Validation Method for Determining the Number of Clusters in Categorical Sequences

Area of Science:

• Computational biology
• Machine learning
• Data mining

Background:

• Cluster validation is crucial for machine learning systems.
• Categorical sequences are increasingly common in real-world applications.
• Existing validation methods are insufficient for sequence data.

Purpose of the Study:

• To address the challenge of validating clusters in categorical sequences.
• To propose a novel cluster validity index (CVI) for sequence data.
• To develop a robust clustering algorithm for categorical sequences.

Main Methods:

• A novel CVI combining intracluster structural compactness and intercluster structural separation.
• A partition-based clustering algorithm with deterministic initialization and noise cluster elimination.
• Integration of the algorithm and CVI for determining the optimal number of clusters.

Main Results:

• Demonstrated effectiveness on protein sequences and real-world datasets.
• The proposed CVI accurately measures the quality of sequence clusters.
• The clustering algorithm yields high-quality results.

Conclusions:

• The novel CVI and clustering algorithm offer a robust solution for categorical sequence validation.
• The method effectively determines the number of clusters in sequence datasets.
• This work advances the field of cluster validation for complex sequence data.