Title | : | Improving the performance of outlier detection methods for categorical data by using weighting function |
Author | : |
Dr. Nur Rokhman, S.Si., M.Kom. (1) Prof. Drs. Subanar, Ph.D. (2) Drs. Edi Winarko, M.Sc.,Ph.D. (3) |
Date | : | 0 2016 |
Keyword | : | Categorical data, Outlier detection, Weighting function Categorical data, Outlier detection, Weighting function |
Abstract | : | Outliers are uncommon events in real life. For a database processing, outlier means unusual records comparing to the remaining records. An outlier can be caused by a damage of a system. A new fact in a system can also cause outliers. Outlier detection is an important task to find an exceptional data. Outlier detection methods for categorical data such as AVF, MR-AVF, AEVF, NAVF, OPAVF, WDOD, and FuzzyAVF work base on the attribute value data frequency. These methods start the outlier detection process by calculating the data of attribute value frequency on each attribute. Then, many complicated calculations based on the various mathematical background are carried out to find the outlier by using the data of attribute value frequency. All the methods above do not take into account on the sparseness of each attribute. In this paper, weighting functions is used to take into account the sparseness of each attribute. AVF and WDOD methods are modified by using weighting function. The performance of these modified methods is observed based on the detected outlier of UCI Machine Learning datasets. The experiment shows that weighting function can improve the performance of AVF and WDOD on the outlier detection in Adult, Mushroom, and Nursery datasets. © 2005 - 2016 JATIT & LLS. All rights reserved. |
Group of Knowledge | : | Ilmu Komputer |
Original Language | : | English |
Level | : | Internasional |
Status | : |
Published
|