Adding Rejection to PyOD

[Lorenzo-Perini]

All Submissions Basics:

• [Y] Have you followed the guidelines in our Contributing document?
• [Y] Have you checked to ensure there aren't other open Pull Requests for the same update/change?
• [Y] Have you checked all Issues to tie the PR to a specific one?

All Submissions Cores:

• [Y] Have you added an explanation of what your changes do and why you'd like us to include them?
• [N] Have you written new tests for your core changes, as applicable?
• [Y] Have you successfully ran tests with your changes locally?
• [Y] Does your submission pass tests, including CircleCI, Travis CI, and AppVeyor?
• [Y] Does your submission have appropriate code coverage? The cutoff threshold is 95% by Coversall.

[Lorenzo-Perini]

Hi,

During NeurIPS 2023, I have presented my paper on Unsupervised Anomaly Detection with Rejection, which allows an anomaly detector to abstain from making highly uncertain predictions. From the code perspective, this means that the model outputs -2 whenever it "abstains".

My opinion is that this could be an interesting feature to add to PyOD, but feel free to reject the pull request if you do not agree. The link to the paper is: https://arxiv.org/pdf/2305.13189.pdf

Best,
Lorenzo

[coveralls]

Pull Request Test Coverage Report for Build 7290418106

• 3 of 44 (6.82%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.4%) to 93.36%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
pyod/models/base.py	3	44	6.82%

Totals
Change from base Build 7233201130:	-0.4%
Covered Lines:	9013
Relevant Lines:	9654

---

💛 - Coveralls