An Algorithm for Clustering Categorical Data With Set-Valued Features.

Area of Science:

• Data Mining and Machine Learning
• Computational Statistics

Background:

• Traditional data mining often assumes single-valued features, limiting analysis of real-world data with multi-valued attributes.
• Existing methods for set-valued features, like dummy coding, can be inefficient and may not accurately represent data characteristics.

Purpose of the Study:

• To propose a novel algorithm, SV-k-modes, for effective clustering of categorical data containing set-valued features.
• To develop a specialized distance function and cluster center representation suitable for set-valued data.

Main Methods:

• The SV-k-modes algorithm utilizes a unique distance metric designed for set-valued features.
• A heuristic method is employed for updating cluster centers iteratively, alongside an initialization algorithm for selecting initial centers.
• Convergence and computational complexity of the proposed algorithm are theoretically analyzed.

Main Results:

• Experimental results on synthetic and real-world datasets demonstrate superior performance of SV-k-modes compared to three other categorical clustering algorithms.
• The algorithm shows significant improvements in clustering accuracy for real data applications.
• SV-k-modes proves to be scalable for handling large datasets.

Conclusions:

• The SV-k-modes algorithm offers an effective solution for clustering categorical data with set-valued features.
• It provides a more accurate and efficient approach than traditional methods, particularly for complex, real-world datasets.
• The algorithm's scalability makes it suitable for large-scale data mining applications.