Biblioteca em Python para explicabilidade de anomalias

Abstract

Outlier detection has been used to detect and, if appropriate, remove anomalous observations from the data. Its usability can identify system failures and frauds before they escalate with potentially huge consequences. One ramification of the contribution made by the outlier detection field is related to understanding which aspects of the anomalous observation significantly separate it from the others in a given dataset. This area of research has been called Outlier Aspect Mining (OAM). Promising results and applications and cases have been presented by the community. With this in mind, the objective of this paper is to contribute to OAM research in a practical way. We developed a Python library that allows the user to apply Outlier Aspect Mining algorithms and analyze the results in their own datasets. The source code is made available to the community not only to use but also to extend as wished. The library was tested on economic market data and also evaluated by an end user.