Elamur Research⌕
☰
home / papers / 2026 / AutoAnoEval: Semantic-Aware Model Selection via Tree-Guided LLM Reasoning for Tabular Anomaly Detection

AutoAnoEval: Semantic-Aware Model Selection via Tree-Guided LLM Reasoning for Tabular Anomaly Detection

Paper · 2026 · Suhee Yoon, Sanghyu Yoon, Ye Seul Sim et al.

Summary

Choosing a tabular anomaly detection (AD) model is hard when no labelled anomalies exist, not even in the evaluation split. Earlier synthetic-anomaly approaches, such as the four anomaly types in ADBench, perturb numerical distributions and ignore what the features actually mean. AutoAnoEval addresses this by building a pseudo evaluation set that combines held-out normal data with LLM-synthesised anomalies grounded in domain semantics.

The pipeline has three steps:

  1. An LLM receives a data sample and a semantic context made of dataset descriptions, feature metadata and statistics. Using chain-of-thought prompting, it writes anomaly conditions in which each value is plausible but the combination is not.
  2. A CART decision tree is repeatedly trained on the current pseudo set. Its normal paths are handed back to the LLM, which extends them into harder anomalies near the decision boundary.
  3. The best iteration is selected by condition diversity, and candidate detectors are ranked on the resulting set.

Across 20 tabular AD benchmarks, model rankings on the pseudo sets closely matched rankings on real anomalies, with small score gaps. This makes model selection workable without collected anomalies.

Key points

  • Generation used Gemini-1.5-Flash, with Gemini-2.0-Pro acting as an independent judge of condition quality.
  • In the context ablation, adding feature-level and then dataset-level context raised average AUROC from 0.755 to 0.789 to 0.809. Average MAE fell from 0.278 to 0.154 to 0.127.
  • Judged condition quality peaked at iterations 2–3 (overall score 0.7209 at iteration 2, 0.6633 at iteration 4). This supports picking the best iteration (ACDS-based) rather than the last one.
  • Using the synthetic anomalies as contrastive negatives during training gave the best average AUROC (0.842) and rank (3.0), ahead of MCM, NeuTraL, SLAD, IForest, AnoLLM and others.
  • Limitations: the method depends on meaningful metadata, and repeated LLM queries plus tree training add compute cost.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Sungkyunkwan UniversitySuwon · South Korea10 works · 195 citations🏢LG AI ResearchSeoul · South Korea8 works · 374 citations

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,685 works

Cite

Cite as Suhee Yoon, Sanghyu Yoon, Ye Seul Sim, Seungdong Yoa, Dongmin Kim, Soonyoung Lee, Hankook Lee and Woohyung Lim. 2026. AutoAnoEval: Semantic-Aware Model Selection via Tree-Guided LLM Reasoning for Tabular Anomaly Detection. Findings of the Association for Computational Linguistics: EACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-eacl.183

@inproceedings{yoon2026autoanoeval,
  title = {{AutoAnoEval: Semantic-Aware Model Selection via Tree-Guided LLM Reasoning for Tabular Anomaly Detection}},
  author = {Suhee Yoon, Sanghyu Yoon, Ye Seul Sim, Seungdong Yoa, Dongmin Kim, Soonyoung Lee, Hankook Lee and Woohyung Lim},
  booktitle = {Findings of the Association for Computational Linguistics: EACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-eacl.183},
  url = {https://doi.org/10.18653/v1/2026.findings-eacl.183}
}

References

39 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Charu C. Aggarwal. 2015. Outlier analysis. In Data Mining, pages 237–263. Springer, Cham. find
  2. Pierre Boyeau, Anastasios N Angelopoulos, Nir Yosef, Jitendra Malik, and Michael I Jordan. 2024. Autoeval done right: Using synthetic data for model evaluation. arXiv preprint arXiv:2403.07008. — [arXiv:2403.07008](https://arxiv.org/abs/2403.07008) find
  3. Markus M Breunig, Hans-Peter Kriegel, Raymond T Ng, and Jörg Sander. 2000. Lof: identifying density-based local outliers. ACM SIGMOD Record, 29(2):93–104. find
  4. Andy Brown, Aaron Tuor, Brian Hutchinson, and Nicole Nichols. 2018. Recurrent neural network attention mechanisms for interpretable system log anomaly detection. In Proceedings of the first workshop on machine learning for computing systems, pages 1–8. find
  5. Fabrizio Carcillo, Yann-Aël Le Borgne, Olivier Caelen, Yacine Kessaci, Frédéric Oblé, and Gianluca Bontempi. 2021. Combining unsupervised and supervised learning in credit card fraud detection. Information sciences, 557:317–331. find
  6. Sihan Chen and 1 others. 2025. Pyod 2: A python library for outlier detection with llm-powered model selection. In Companion Proceedings of the ACM on Web Conference 2025. ACM. find
  7. Edward Choi, Mohammad Taha Bahadori, Jimeng Sun, Joshua Kulas, Andy Schuetz, and Walter Stewart. 2016. Retain: An interpretable predictive model for healthcare using reverse time attention mechanism. find
  8. Xueying Ding, Yue Zhao, and Leman Akoglu. 2024. Fast unsupervised deep outlier model selection with hypernetworks. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. ACM. find
  9. Tuan Dinh, Yuchen Zeng, Ruisu Zhang, Ziqian Lin, Michael Gira, Shashank Rajput, Jy-yong Sohn, Dimitris Papailiopoulos, and Kangwook Lee. 2022. Lift: Language-interfaced fine-tuning for non-language machine learning tasks. In Advances in Neural Information Processing Systems, volume 35, pages 11763–… find
  10. Markus Goldstein and Andreas Dengel. 2012. Histogram-based outlier score (hbos): A fast unsupervised anomaly detection algorithm. In KI-2012: Poster and Demo Track, pages 59–63. find
  11. Alex Goodge, Bryan Hooi, See Kiong Ng, and Wee Siong Ng. 2022. Lunar: Unifying local outlier detection methods via graph neural networks. In Proceedings of the AAAI Conference on Artificial Intelligence. AAAI. find
  12. S. Han, J. Yoon, S. O. Arik, and T. Pfister. 2024. Large language models can automatically engineer features for few-shot tabular learning. arXiv preprint arXiv:2404.09491. — [arXiv:2404.09491](https://arxiv.org/abs/2404.09491) find
  13. Songqiao Han and 1 others. 2022. Adbench: Anomaly detection benchmark. In Advances in Neural Information Processing Systems, volume 35, pages 32142– 32159. find
  14. Johanna Hardin and David M Rocke. 2004. Outlier detection in the multiple cluster setting using the minimum covariance determinant estimator. Computational Statistics & Data Analysis, 44(4):625–638. find
  15. Zengyou He, Xiaofei Xu, and Shengchun Deng. 2003. Discovering cluster-based local outliers. Pattern Recognition Letters, 24(9-10):1641–1650. find
  16. Stefan Hegselmann, Alejandro Buendia, Hunter Lang, Monica Agrawal, Xiaoyi Jiang, and David Sontag. 2023. Tabllm: Few-shot classification of tabular data with large language models. In Proceedings of the International Conference on Artificial Intelligence and Statistics, pages 5549–5581. PMLR. find
  17. Andrey Kharitonov, Abdulrahman Nahhas, Matthias Pohl, and Klaus Turowski. 2022. Comparative analysis of machine learning models for anomaly detection in manufacturing. Procedia Computer Science, 200:1288–1297. find
  18. Alon Shoshan, Nadav Bhonker, Igor Kviatkovsky, Matan Fintz, and Gérard Medioni. 2023. Synthetic data for model selection. In Proceedings of the International Conference on Machine Learning, pages 31633–31656. PMLR. find
  19. Hans-Peter Kriegel and Arthur Zimek. 2008. Angle-based outlier detection in high-dimensional data. In Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD ’08), pages 444–452. ACM. find
  20. Mei-Ling Shyu, Shu-Ching Chen, Kanoksri Sarinna-pakorn, and Li Chang. 2003. A novel anomaly detection scheme based on principal component classifier. Technical report, Department of Electrical and Computer Engineering, University of Miami, Coral Gables, FL. find
  21. Zheng Li, Yue Zhao, Nicola Botta, Cezar Ionescu, and Xiyang Hu. 2020. Copod: Copula-based outlier detection. In IEEE International Conference on Data Mining (ICDM). IEEE. find
  22. Fei Tony Liu, Kai Ming Ting, and Zhi-Hua Zhou. 2008. Isolation forest. In IEEE International Conference on Data Mining, pages 413–422. IEEE. find
  23. Yue Liu, Zheng Li, Chenglin Zhou, Yujia Jiang, Jing Sun, Min Wang, and Xiangnan He. 2019. Generative adversarial active learning for unsupervised outlier detection. IEEE Transactions on Knowledge and Data Engineering. find
  24. Pankaj Malhotra, Anusha Ramakrishnan, Gaurangi Anand, Lovekesh Vig, Puneet Agarwal, and Gautam Shroff. 2016. Lstm-based encoder-decoder for multi-sensor anomaly detection. arXiv preprint arXiv:1607.00148. — [arXiv:1607.00148](https://arxiv.org/abs/1607.00148) find
  25. J. Nam, K. Kim, S. Oh, J. Tack, J. Kim, and J. Shin. 2024. Optimized feature generation for tabular data via llms with decision tree reasoning. In Advances in Neural Information Processing Systems, volume 37, pages 92352–92380. find
  26. Tomáš Pevný. 2016. Loda: Lightweight on-line detector of anomalies. Machine Learning, 102(2):275–304. find
  27. Sanjay Purushotham, Chuizheng Meng, Zhengping Che, and Yan Liu. 2018. Benchmarking deep learning models on large healthcare datasets. Journal of biomedical informatics, 83:112–134. find
  28. Suresh Ramaswamy, Rajeev Rastogi, and Kyuseok Shim. 2000. Efficient algorithms for mining outliers from large data sets. ACM SIGMOD Record, 29(2):427–438. find
  29. Lukas Ruff, Robert Vandermeulen, Nico Goernitz, Lucas Deecke, Shoaib A. Siddiqui, Alexander Binder, Emmanuel Müller, and Marius Kloft. 2018. Deep one-class classification. In Proceedings of the International Conference on Machine Learning (ICML), pages 4393–4402. PMLR. find
  30. Marco Schreyer, Timur Sattarov, Christian Schulze, Bernd Reimer, and Damian Borth. 2019. Detection of accounting anomalies in the latent space using adversarial autoencoder neural networks. arXiv preprint arXiv:1908.00734. — [arXiv:1908.00734](https://arxiv.org/abs/1908.00734) find
  31. Jian Tang, Zhixiang Chen, Ada Wai-Chee Fu, and David Wai-Lok Cheung. 2002. Enhancing effectiveness of outlier detections for low density patterns. In Pacific-Asia Conference on Knowledge Discovery and Data Mining, pages 535–548, Berlin, Heidelberg. Springer. find
  32. Che-Ping Tsai, Ganyu Teng, Phillip Wallis, and Wei Ding. 2025. AnoLLM: Large language models for tabular anomaly detection. In The Thirteenth International Conference on Learning Representations. find
  33. Boris van Breugel, Nabeel Seedat, Fergus Imrie, and Mihaela van der Schaar. 2023. Can you rely on your model evaluation? improving model evaluation with synthetic test data. In Advances in Neural Information Processing Systems, volume 36, pages 1889– 1904. find
  34. Haowen Xu, Wenxiao Chen, Nengwen Zhao, Zeyan Li, Jiahao Bu, Zhihan Li, Ying Liu, Youjian Zhao, Dan Pei, Yang Feng, and 1 others. 2018. Unsupervised anomaly detection via variational auto-encoder for seasonal kpis in web applications. In Proceedings of the 2018 world wide web conference, pages 187–19… find
  35. Tiankai Yang and 1 others. 2025. Ad-agent: A multi-agent framework for end-to-end anomaly detection. arXiv preprint arXiv:2505.12594. find
  36. Sanghyu Yoon, Dongmin Kim, Suhee Yoon, Ye Seul Sim, Seungdong Yoa, Hye-Seung Cho, Soonyoung Lee, Hankook Lee, and Woohyung Lim. 2025. Retabad: A benchmark for restoring semantic context in tabular anomaly detection. arXiv preprint arXiv:2510.02060. — [arXiv:2510.02060](https://arxiv.org/abs/2510.02060) find
  37. Houssam Zenati, Manon Romain, Chuan Sheng Foo, Benjamin Lecouat, and Vijay Chandrasekhar. 2018. Adversarially learned anomaly detection. In 2018 IEEE International Conference on Data Mining (ICDM), pages 727–736. IEEE. find
  38. Yue Zhao, Ryan Rossi, and Leman Akoglu. 2021. Automatic unsupervised outlier model selection. In Thirty-Fifth Conference on Neural Information Processing Systems. find
  39. Yue Zhao, Sean Zhang, and Leman Akoglu. 2022. Toward unsupervised outlier model selection. In 2022 IEEE International Conference on Data Mining (ICDM), page in press. IEEE. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yoon-2026-autoanoeval-semantic-aware-model-selection/