@inproceedings{vinci2025Rakuschek,
author = {Rakuschek, Julian and Leitner, Michael and Bernard, J\"{u}rgen and Wriessnegger, Selina Christin and Schreck, Tobias},
title = {AnoScout - Visual Exploration of Anomalies and Anomaly Detection Algorithm Ensembles in Time Series Data},
year = {2025},
isbn = {9798400718458},
publisher = {ACM},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3769534.3769577},
pdf = {https://dl.acm.org/doi/pdf/10.1145/3769534.3769577},
doi = {10.1145/3769534.3769577},
abstract = {With the growing abundance of time series data and anomaly detection algorithms, selecting appropriate algorithm configurations for a given dataset has become increasingly complex. We introduce AnoScout, a Visual Analytics approach to explore anomalies obtained from an algorithm ensemble with the overall goal of acquiring insights into the diversity of anomalies and identifying appropriate algorithms for each anomaly pattern. We employ unsupervised methods1 to address scenarios in which normal behavior is difficult to define, and integrate semi-supervised approaches with projection-based visualizations to support user labeling when normal behavior can be more clearly delineated. Our approach considers ensembles of algorithms, enabling robust coverage across multiple anomaly categories. AnoScout visualizes each algorithm’s contribution to the ensemble to address the challenge of differing detection behaviors across anomaly types. To support analysis in large datasets with potentially many anomalies, we integrate a recommender system that facilitates the identification of relevant anomalies. To acquire insights into recurring anomalies, users can explore anomalies through a clustering view. We demonstrate the practical utility of AnoScout using case studies from two domains: EEG measurements and industrial data analysis, which are known for containing diverse anomalies.},
booktitle = {Visual Information Communication and Interaction},
articleno = {9},
numpages = {5},
keywords = {Time Series, Anomaly Detection, Ensembles, Visual Analytics},
teaserpage = {1},
topics = {Visual Analytics for Time-Oriented Data, Interactive Machine Learning},
code = {P032},