eif | Extended Isolation Forest for Anomaly Detection | Predictive Analytics library

eif is a Jupyter Notebook library typically used in Analytics, Predictive Analytics applications. eif has no bugs, it has no vulnerabilities and it has low support. However eif has a Non-SPDX License. You can download it from GitHub.

The problem of anomaly detection has wide range of applications in various fields and scientific applications. Anomalous data can have as much scientific value as normal data or in some cases even more, and it is of vital importance to have robust, fast and reliable algorithms to detect and flag such anomalies. Here, we present an extension to the model-free anomaly detection algorithm, Isolation Forest Liu2008. This extension, named Extended Isolation Forest (EIF), improves the consistency and reliability of the anomaly score produced by standard methods for a given data point. We show that the standard Isolation Forest produces inconsistent anomaly score maps, and that these score maps suffer from an artifact produced as a result of how the criteria for branching operation of the binary tree is selected. Our method allows for the slicing of the data to be done using hyperplanes with random slopes which results in improved score maps. The consistency and reliability of the algorithm is much improved using this extension. Here we show the need for an improvement on the source algorithm to improve the scoring of anomalies and the robustness of the score maps especially around edges of nominal data. We discuss the sources of the problem, and we present an efficient way for choosing these hyperplanes which give way to multiple extension levels in the case of higher dimensional data. The standard Isolation Forest is therefore a special case of the Extended Isolation Forest as presented it here. For an N dimensional dataset, Extended Isolation Forest has N levels of extension, with 0 being identical to the case of standard Isolation Forest, and N-1 being the fully extended version.