Review of algorithms for the detection of outliers

Abstract

The detection of outliers is an extremely important task in a wide variety of application domains.  Often these values are eliminated to improve the accuracy of the information, but sometimes the presence of an outlier has a certain sense or explanation that can be lost if it can be eliminated, that its identification can lead to the discovery of unexpected knowledge.  Various areas such as:   criminal activities in electronic commerce, fraud detection and even statistical performance analysis.  The article presented is the result of a non-exhaustive documentary investigation of the opinion of several authors, who focused their work to determine the efficiency of the methods or algorithms for the detection of outliers.  Initially, a theoretical conceptual study was carried out to understand the nature of an atypical value and its classification, and then perform an analysis on the different techniques in the determination of clusters, distances and density. For each one of the techniques of detection of atypical values, it was found that the algorithms that have been created by different authors besides the efficiency that each of them has in certain contexts.