Outliers Identification Model in Point-of-Sales Data Using Enhanced Normal Distribution Method
Fahed Yoseph, Markku Heikkilä, Daniel Howard
Abstract
Fahed Yoseph, Markku Heikkilä, Daniel Howard
Abstract
Data Mining extrapolates patterns drawing conclusions from data. Outliers detection identifies those objects that fall some standard deviations away from the mean and is an important tool of commercial data mining. Characterizing the manner of outliers can lead to new knowledge, such as the manner of fraudulent transactions. However, outliers may represent meaningless aberrations and hence there is no rigid mathematical or statistical definition of what constitutes an outlier, and, in many scenarios, determination of the outlier is ultimately a subjective exercise. Standard deviation is a central actor in outlier detection and yet exhibits sensitivity to values and can be distorted, inflated, by a single or even a few observations of borderline and extreme values. It can mask the situation where less extreme outliers or anomalies go undetected because of the existence of the most extreme outliers. This study proposes a novel outlier identification model using an enhanced normal distribution method. The model can explore different types of outliers giving an end-user the ability to fully or partially eliminate outliers found in a retail point of sale (POS) dataset. Experiments revealed that the enhanced normal distribution method appeared more accurate than the standard normal distribution method, and results were also evaluated subjectively by the client, who found most of the outliers to be truly outliers and some representing potentially fraudulent transactions.
OpenAlex reports 8 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
Data Mining extrapolates patterns drawing conclusions from data. Outliers detection identifies those objects that fall some standard deviations away from the mean and is an important tool of commercial data mining. Characterizing the manner of outliers can lead to new knowledge, such as the manner of fraudulent transactions. However, outliers may represent meaningless aberrations and hence there is no rigid mathematical or statistical definition of what constitutes an outlier, and, in many scenarios, determination of the outlier is ultimately a subjective exercise. Standard deviation is a central actor in outlier detection and yet exhibits sensitivity to values and can be distorted, inflated, by a single or even a few observations of borderline and extreme values. It can mask the situation where less extreme outliers or anomalies go undetected because of the existence of the most extreme outliers. This study proposes a novel outlier identification model using an enhanced normal distribution method. The model can explore different types of outliers giving an end-user the ability to fully or partially eliminate outliers found in a retail point of sale (POS) dataset. Experiments revealed that the enhanced normal distribution method appeared more accurate than the standard normal distribution method, and results were also evaluated subjectively by the client, who found most of the outliers to be truly outliers and some representing potentially fraudulent transactions.
Key concepts: Outlier, Anomaly detection, Computer science, Identification (biology), Standard deviation, Data mining, Normal distribution, Point (geometry)