Affiliations: Data Mining and Optimization Research Group, Center for Artificial Intelligence Technology, Faculty of Information Science and Technology, University Kebangsaan Malaysia, Bangi, Selangor Darul Ehsan, Malaysia
Correspondence:
[*]
Corresponding author: Djoko Budiyanto Setyohadi, Data Mining and Optimization Research Group, Center for Artificial Intelligence Technology, Faculty of Information Science and Technology, University Kebangsaan Malaysia, Bangi, Selangor Darul Ehsan 43000, Malaysia. E-mail:djokobdy@gmail.com
Abstract: Technically, the problem of overlap in a dataset is viewed as an uncertainty
problem and is solved using a fuzzy set theoretical approach, specifically,
fuzzy clustering. This approach is powerful but has some problems associated
with it, of which the design of the membership function is the most serious.
There are many different techniques for optimizing fuzzy clustering,
including those based on similarity decomposition and centroids of clusters.
Furthermore, the problem of overlap clustering is still being studied to
improve its performance, especially with respect to the membership
optimization. Rough set theory (RST) is the complement of fuzzy set theory
and evidence theory, which use different techniques to address the
uncertainty problem in overlap clustering. Considering the simplicity of the
membership computation in RST, we propose an overlap clustering algorithm,
which involves the use of the discernibility concept of RST to improve the
overlap clusters as an existing variant of the overlap clustering algorithm.
The experiment described here demonstrates that this new method improves the
performance and increases the accuracy of clustering while avoiding the time
complexity problem. The experiment uses five UCI machine learning datasets.
The complexity of the data is measured using the volume of the overlap
region and feature efficiency. The experimental results show that the
proposed method significantly outperforms the other two methods in terms of
the Dunn index, the sum of the squared errors and the silhouette index.
