UNIVERSITAS BINA DARMA, UNIVERSITAS BINA DARMA (2022) ATTRIBUTE VALUE PAIRS BASED ON DISCERNIBILITY MATRIX FOR OUTLIERS DETECTION. ATTRIBUTE VALUE PAIRS BASED ON DISCERNIBILITY MATRIX FOR OUTLIERS DETECTION.
|
Text
ATTRIBUTE VALUE PAIRS BASED ON DISCERNIBILITY MATRIX FOR OUTLIERS DETECTION.pdf Download (1MB) | Preview |
Abstract
Outlier mining is one important task in data mining and it has always been receiving attention from many researchers. The detection of outliers is found useful in many real applications like fraud detection and network intrusion. There are many outlier detection methods found in literature which include the frequent pattern generation and Rough Set based outlier detection. Although many methods have been proposed in data mining, the problems in detecting outliers efficiently continue especially in many real applications, due to the high dimensionality of huge data sets and high computational in processing. In this study, we proposed a method to detect outliers by discovering interesting attribute value pairs based on the Discernibility Attribute Value Matrix (DAV) in Rough Set Theory (RS). Interesting attribute value pairs (avp) are generated from the DAV Matrix. Two measures which are the support and interest value are used to measure the interestingness of the attributes. In order to detect outliers, a new measurement called the DAV Outlier factor (DAVOF) is proposed. In addition, an Average Ratio (AR), which measures the performance of the outlier detection method is also proposed. The DAV algorithm (DAVAlg) is compared with the FindFPOF and RSetAlg methods. The result shows that the DAVAlg outperforms the other two methods.
| Item Type: | Article |
|---|---|
| Subjects: | H Social Sciences > H Social Sciences (General) |
| Divisions: | Faculty of Law, Arts and Social Sciences > School of Social Sciences |
| Depositing User: | Mr Edi Surya Negara |
| Date Deposited: | 24 Jun 2022 15:23 |
| Last Modified: | 24 Jun 2022 15:23 |
| URI: | http://eprints.binadarma.ac.id/id/eprint/15555 |
Actions (login required)
 |
View Item |
