Space Structure and Clustering of Categorical Data

Area of Science:

• Machine learning
• Data mining
• Pattern recognition

Background:

• Categorical data clustering is crucial for machine learning and data mining.
• Existing k-modes algorithms show good performance but have limitations compared to numeric data clustering.
• Categorical data lacks the clear spatial structure found in numeric data, hindering clustering effectiveness.

Purpose of the Study:

• To propose a novel data representation scheme for categorical data.
• To develop a general framework for space structure-based categorical clustering (SBC) algorithms.
• To enhance the performance of categorical clustering by leveraging Euclidean space mapping.

Main Methods:

• Developed a novel data-representation scheme mapping categorical objects into a Euclidean space.
• Designed a general framework for space structure-based categorical clustering (SBC) algorithms.
• Implemented two versions of SBC algorithms using different dissimilarity measures and compared them with k-modes algorithms.

Main Results:

• The proposed SBC-type algorithms significantly outperform representative k-modes algorithms.
• Experiments demonstrate the effectiveness of mapping categorical data into Euclidean space for improved clustering.
• The SBC framework provides a robust approach for categorical data analysis.

Conclusions:

• The novel data representation and SBC framework effectively address limitations in categorical data clustering.
• SBC algorithms offer superior performance compared to existing k-modes methods.
• This approach opens new avenues for advanced categorical data analysis and knowledge discovery.