Attribute Set Weighting and Decomposition Approaches for Reduct Computation
This research is mainly in the Rough Set theory based knowledge reduction for data classification within the data mining framework. To facilitate the Rough Set based classification, two main knowledge reduction models are proposed. The first model is an approximate approach for object reducts compu...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2005
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/5853/1/FSKTM_2005_7%20IR.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This research is mainly in the Rough Set theory based knowledge reduction for data classification within the data mining framework. To facilitate the Rough Set based
classification, two main knowledge reduction models are proposed. The first model is an approximate approach for object reducts computation used particularly for the
data classification purposes. This approach emphasizes on assigning weights for each attribute in the attributes set. The weights give indication for the importance of an
attribute to be considered in the reduct. This proposed approach is named Object Reduct by Attribute Weighting (ORAW). A variation of this approach is proposed to
compute full reduct and named Full Reduct by Attribute Weighting (FRAW).The second proposed approach deals with large datasets particularly with large number of attributes. This approach utilizes the principle of incremental attribute set decomposition to generate an approximate reduct to represent the entire dataset. This
proposed approach is termed for Reduct by Attribute Set Decomposition (RASD).The proposed reduct computation approaches are extensively experimented and
evaluated. The evaluation is mainly in two folds: first is to evaluate the proposed
approaches as Rough Set based methods where the classification accuracy is used as
an evaluation measure. The well known IO-fold cross validation method is used to
estimate the classification accuracy. The second fold is to evaluate the approaches as
knowledge reduction methods where the size of the reduct is used as a reduction
measure. The approaches are compared to other reduct computation methods and to other none Rough Set based classification methods. The proposed approaches are applied to various standard domains datasets from the UCI repository. The results of the experiments showed a very good performance for the proposed approaches as classification methods and as knowledge reduction methods. The accuracy of the ORAW approach outperformed the Johnson approach over all the datasets. It also produces better accuracy over the Exhaustive and the Standard Integer Programming (SIP) approaches for the majority of the datasets used in the experiments. For the RASD approach, it is compared to other classification methods and it shows very competitive results in term of classification accuracy and reducts size. As a conclusion, the proposed approaches have shown competitive and even better accuracy in most tested domains. The experiment results indicate that the proposed approaches as Rough classifiers give good performance across different classification problems and they can be promising methods in solving classification problems. Moreover, the experiments proved that the incremental vertical decomposition framework is an appealing method for knowledge reduction over large datasets within the framework of Rough Set based classification. |
|---|