Weighted Kernel based Prediction and Detection of Outliers
The topic of finding and understanding anomalous values in categorizing databases, has been analysed in this paper. An attribute value is considered to be unusual if its frequency is out of the ordinary within the general frequencies and percentages. As the first major addition, the concept of frequ...
Saved in:
Published in | 2022 3rd International Conference on Smart Electronics and Communication (ICOSEC) pp. 219 - 224 |
---|---|
Main Authors | , , , , , |
Format | Conference Proceeding |
Language | English |
Published |
IEEE
20.10.2022
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | The topic of finding and understanding anomalous values in categorizing databases, has been analysed in this paper. An attribute value is considered to be unusual if its frequency is out of the ordinary within the general frequencies and percentages. As the first major addition, the concept of frequent occurrence is introduced. This metric may be viewed as a variant of Kernel Density Estimation adapted to the frequencies. Additionally, an outlierness measurement is constructed for categorical data based on the frequency occurrences distribution's cumulated frequency tables. This metric can detect two types of anomalies: lower and higher outliers which correspond to lower or higher frequency values. In addition, interpretable explanations for unusual data values are given. This work emphasises that giving open to interpretation for the information extracted is a desired component of any knowledge extraction strategy, despite the fact that most classic outlier identification methods do not. |
---|---|
DOI: | 10.1109/ICOSEC54921.2022.9952112 |