OAB - An Open Anomaly Benchmark Framework for Unsupervised and Semisupervised Anomaly Detection on Image and Tabular Data Sets

We introduce OAB, an Open Anomaly Benchmark Framework for unsupervised and semisupervised anomaly detection on image and tabular data sets, ensuring simple reproducibility for existing benchmark results as well as a reliable comparability and low-effort extensibility when new anomaly detection algor...

Full description

Saved in:
Bibliographic Details
Published in2021 International Conference on Data Mining Workshops (ICDMW) pp. 991 - 1000
Main Authors Lohrer, Andreas, Deller, Jan, Hunemorder, Maximilian, Kroger, Peer
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.12.2021
Subjects
Anomaly detection
Benchmark
Benchmark testing
Datasets
Evaluation
Machine learning
Machine learning algorithms
Reliability
Reproducibility
Reproducibility of results
Task analysis
Unsupervised and Semisupervised Anomaly Detection
Online AccessGet full text

Cover

More Information
Summary:We introduce OAB, an Open Anomaly Benchmark Framework for unsupervised and semisupervised anomaly detection on image and tabular data sets, ensuring simple reproducibility for existing benchmark results as well as a reliable comparability and low-effort extensibility when new anomaly detection algorithms or new data sets are added. While making established methods of the most popular benchmarks easily accessible, OAB generalizes the task of un- and semisupervised anomaly benchmarking and offers besides commonly used benchmark data sets also semantically meaningful real-world anomaly data sets as well as a broad range of traditional and state-of-the-art anomaly detection algorithms. The benefit of OAB for the research community has been demonstrated by reproducing and extending existing benchmarks to new algorithms with very low effort allowing researchers to focus on the actual algorithm research.
ISSN:2375-9259
DOI:10.1109/ICDMW53433.2021.00129