ReTabAD: A Benchmark for Restoring Semantic Context in Tabular Anomaly Detection

Sanghyu Yoon, Dongmin Kim, Suhee Yoon, Ye Seul Sim, Seungdong Yoa,
Hye-Seung Cho, Soonyoung Lee, Hankook Lee, Woohyung Lim

📄 Paper GitHub 🗂 Dataset

🎯 Overview

ReTabAD Overview
In tabular anomaly detection (AD), textual semantics often carry critical signals, as the definition of an anomaly is closely tied to domain-specific context. However, existing benchmarks provide only raw data points without semantic context, overlooking rich textual metadata such as feature descriptions and domain knowledge that experts rely on in practice. This limitation restricts research flexibility and prevents models from fully leveraging domain knowledge for detection. ReTabAD addresses this gap by Restoring textual semantics to enable context-aware Tabular AD research. We provide (1) 20 carefully curated tabular datasets enriched with structured textual metadata, together with implementations of state-of-the-art AD algorithms—including classical, deep learning, and LLM-based approaches—and (2) a zero-shot LLM framework that leverages semantic context without task-specific training, establishing a strong baseline for future research. Furthermore, this work provides insights into the role and utility of textual metadata in AD through experiments and analysis. Results show that semantic context improves detection performance and enhances interpretability by supporting domain-aware reasoning. These findings establish ReTabAD as a benchmark for systematic exploration of context-aware AD.

✨ Key Features

📚 Semantically-Rich Datasets

Tabular data paired with comprehensive JSON text metadata containing column descriptions, logical types, and characterizations of normal data.

💡 Support SOTA Algorithms

Unified pipeline enabling fair comparisons across traditional ML, deep learning, and modern LLM approaches.

🚀 LLM Potential

Demonstrates substantial performance improvements when models can leverage semantic information.

🔬 Why ReTabAD?

Traditional tabular AD benchmarks exhibit a fundamental disconnect from industrial practice:

ReTabAD solves these problems by restoring semantic context and enabling context-aware AD research.

📊 Benchmark Statistics

ReTabAD includes 20 diverse datasets spanning multiple domains:

Dataset Name Datapoints Columns Normal Count Anomaly Count Anomaly Ratio (%)
automobile 159 25 117 42 26.42
backdoor 29,223 42 29,113 110 0.38
campaign 7,842 16 6,056 1,786 22.77
cardiotocography 2,126 21 1,655 471 22.15
census 50,000 41 47,121 2,879 5.76
churn 7,032 19 5,163 1,869 26.58
cirrhosis 247 17 165 82 33.20
covertype 50,000 12 49,520 480 0.96
credit 30,000 23 23,364 6,636 22.12
equip 7,672 6 6,905 767 10.00
gallstone 241 38 161 80 33.20
glass 214 9 163 51 23.83
glioma 730 23 487 243 33.29
quasar 50,000 8 40,520 9,480 18.96
seismic 2,584 18 2,414 170 6.58
stroke 4,909 10 4,700 209 4.26
vertebral 310 6 210 100 32.26
wbc 535 30 357 178 33.27
wine 178 13 130 48 26.97
yeast 1,484 8 1,389 95 6.40

🚀 Quick Start

# Clone the repository
git clone https://github.com/yoonsanghyu/ReTabAD.git
cd ReTabAD

# Build Docker image
docker build -t retabad:1.0.0 .

# Run experiment
python run_default.py --data_name wine --model_name OCSVM --cfg_file configs/default/pyod/OCSVM.yaml

See Usage for detailed instructions.

📰 News