Rough set approach for categorical data clustering

• Main Author: Herawan, Tutut
• Format: Thesis
• Language: English; English; English
• Published: 2010
• Subjects: QA Mathematics; QA71-90 Instruments and machines
• Online Access: http://eprints.uthm.edu.my/3609/1/24p%20TUTUT%20HERAWAN.pdf; http://eprints.uthm.edu.my/3609/2/TUTUT%20HERAWAN%20COPYRIGHT%20DECLARATION.pdf; http://eprints.uthm.edu.my/3609/3/TUTUT%20HERAWAN%20WATERMARK.pdf; http://eprints.uthm.edu.my/3609/
• Institution: Universiti Tun Hussein Onn Malaysia

A few techniques of rough categorical data clustering exist to group objects having similar characteristics. However, the performance of the techniques is an issue due to low accuracy, high computational complexity and clusters purity. This work proposes a new technique called Maximum Dependency Attributes (MDA) to improve the previous techniques due to these issues. The proposed technique is based on rough set theory by taking into account the dependency of attributes of an information system. The main contribution of this technique is to introduce a new technique to classify objects from categorical datasets which has better performance as compared to the baseline techniques. The algorithm of the proposed technique is implemented in MATLAB® version 7.6.0.324 (R2008a). They are executed sequentially on a processor Intel Core 2 Duo CPUs. The total main memory is 1 Gigabyte and the operating system is Windows XP Professional SP3. Results collected during the experiments on four small datasets and thirteen UCI benchmark datasets for selecting a clustering attribute show that the proposed MDA technique is an efficient approach in terms of accuracy and computational complexity as compared to BC, TR and MMR techniques. For the clusters purity, the results on Soybean and Zoo datasets show that MDA technique provided better purity up to 17% and 9%, respectively. The experimental result on supplier chain management clustering also demonstrates how MDA technique can contribute to practical system and establish the better performance for computation complexity and clusters purity up to 90% and 23%, respectively.