Efficient layered density-based clustering of categorical data.

Area of Science:

• Bioinformatics
• Data Mining
• Machine Learning

Background:

• Density-based clustering of categorical biomedical data faces challenges due to undefined attribute value ordering, slowing dense subspace searches.
• Existing methods struggle with scalability and efficiency on large, dynamic datasets.

Purpose of the Study:

• To introduce the HIERDENC algorithm for hierarchical density-based clustering of categorical data.
• To develop an efficient indexing method for fast dense subspace searching.
• To address the limitations of existing clustering algorithms in handling large and evolving biomedical datasets.

Main Methods:

• Developed the HIERDENC algorithm for hierarchical density-based clustering of categorical data.
• Proposed a complementary index for efficient dense subspace searching.
• Implemented an updating mechanism for the HIERDENC index to accommodate new objects without full re-clustering.
• Performed comparisons with other clustering algorithms on large datasets.

Main Results:

• HIERDENC demonstrated superior runtime scalability with increasing numbers of objects compared to other algorithms.
• Achieved improved cluster quality on large datasets.
• Successfully collapsed bicliques in large networks, resulting in up to an 86.5% edge reduction.
• The HIERDENC index updates and cluster retrieval were found to be efficient.

Conclusions:

• HIERDENC is a scalable and efficient algorithm for hierarchical density-based clustering of categorical data.
• The algorithm is well-suited for large and rapidly growing datasets due to its independence from object ordering and efficient incremental updating.
• HIERDENC requires no user-specified input parameters, simplifying its application.