Talk:Isolation forest

Computational complexity of Isolation Forest

It very unlikely that Isolation Forest (IF) is the world fastest anomaly detection algorithm. In essence, IF is at least O(n), and will likely be much slower than simple statistical detectors such as the following: 1: Compute empirical mean and standard deviation, 2: Label all observations as anomalies that violate the 3 sigma rule. — Preceding unsigned comment added by Mbt19 (talk • contribs) 08:35, 26 November 2021 (UTC)

indeed, there is the proposal to just subsample a few objects, and use the distance to the closest as anomaly score: Mahito Sugiyama, Karsten M. Borgwardt:

Rapid Distance-Based Outlier Detection via Sampling. NIPS 2013: 467-475. I bet this is usually faster. 93.132.170.165 (talk) 22:05, 6 July 2022 (UTC)

Poor introductory example

The introductory sample with the web server in "Fig. 1" is misleading. This is longitudinal data, but isolation forest is for point data, ignoring time. 93.132.170.165 (talk) 22:05, 6 July 2022 (UTC)

This is not true! Isolation Forest is frequently used in anomaly detection in time series data: there are several major projects in Kaggle that demonstrate that, e.g. https://www.kaggle.com/code/ksmooi/anomaly-detection-in-time-series-isolationforest. Sal Borrelli (talk) 15:28, 4 February 2026 (UTC)

I added a new example, please check if this is appropriate.Mbt19 (talk) 10:37, 8 July 2022 (UTC)

Repeated self-promotion, cite spam, and inaccurate content

It appears that a primary contributor has WP:ACTUALCOI and is unfortunately not improving the quality from a technical point of view. I am seeing unnecessary references being included (both 2008 publications of Isolation forest: The conference AND the journal publication whose content is very similar), self-promotion (the name of the primary author of isolation forest is explicitly mentioned in the lead text, the history section AND in "Open Source Implementation" section), removal of perfectly fine technical content by other authors, as well as nonsensical promotion of the method itself. For example, in the lead section the author added Isolation Forest is fast because it splits the data space randomly, using randomly selected attribute and randomly selected split point. This is nonsensical, because random splitting has no relation with an algorithm's execution speed. The true reason why isolation forest CAN BE fast (if O(1) number of trees are built) is because each split is binary and hence only logarithmically many splits are needed to build a full tree. Also No density estimation is performed in the algorithm is a pointless statement, because no tree-based data mining algorithm does that. The entire lead section does not really make sense anymore since it focuses on distinguishing isolation forest from other data mining methods rather than describing what the method is and how it works. Finally, statements such as Isolation Forest is fast, ... is the world fastest anomaly detector have been repeatedly added to the article. Isolation forest is NOT fast. Its runtime is linear with a moderate constant overhead that scales with the number of trees. Whether linear runtime is "fast" depends on the context and cannot be claimed in general. Other anomaly detectors with linear runtime can have much less constant overhead, e.g., copula-based outlier detection when the data are low-dimensional.Mbt19 (talk) 18:34, 12 October 2023 (UTC)

This is a recurring problem with small articles on obscure algorithms. Feel free to strip out anything that is inaccurate, poorly sourced, or overly promotional. Suriname0 (talk) 19:51, 12 October 2023 (UTC)

You are right re: ACTUALCOI and for this reason I (the original author) reverted the article to a previous release. Please send me a message if you find any further inconsistencies like that.

Thank you! Sal Borrelli (talk) 15:51, 4 February 2026 (UTC)

Wiki Education assignment: Research Process and Methodology - SP25 - Sect 202 - Thu

This article was the subject of a Wiki Education Foundation-supported course assignment, between 22 January 2025 and 30 April 2025. Further details are available on the course page. Student editor(s): Kj2578 (article contribs).

— Assignment last updated by Kj2578 (talk) 23:57, 22 March 2025 (UTC)

Reverted to previous version

I reverted the article to a previous release, because this article has been the target of scammers, persons with conflicts of interests, subjective opinions, etc. If you have any idea on this article, please post your comments here and we can discuss; otherwise, please simply shut up and do not try anymore to take ownership of other peoples' hard work with your useless edits.

Next time I find a scammer editing my and other serious people's work I will report your account to Wikipedia and request an immediate cancellation. If you insist, I will report you to the nearest cybercrime police department. Sal Borrelli (talk) 15:50, 4 February 2026 (UTC)

Badly reference 'history' part

I noticed that the History subpart was badly referenced. It mentions amelioration in an article published in 2012 and the citation is from a paper published in 2009. Also, it mentions a paper by similar authors from the source Isolation Forest paper and it is from a different theme. This is misleading. Sittelle 21 (talk) 06:32, 16 April 2026 (UTC)