MacrOData: New Benchmarks of Thousands of Datasets for Tabular Outlier Detection

A research team has released MacrOData, a benchmark suite for tabular outlier detection that spans 2,446 datasets, dwarfing the 57-dataset AdBench standard and providing standardized splits and a public leaderboard [1][2]. The suite, detailed in a paper posted to the arXiv preprint repository, is organized into three components: OddBench, OvrBench, and SynBench [1][2]. OddBench contains 790 datasets with real-world semantic anomalies, OvrBench holds 856 datasets featuring real-world statistical outliers, and SynBench supplies 800 synthetically generated datasets that vary data priors and outlier archetypes [2]. The authors argue that the small scale of AdBench, long the de facto benchmark in the field, has restricted diversity and statistical power [2]. All 2,446 datasets are open-sourced, and the project includes a publicly accessible leaderboard hosted on Hugging Face [2]. The benchmark provides standardized train/test splits for every dataset and annotates each with semantic metadata [2]. A portion of the test labels is held out for a private partition that feeds the online leaderboard, a design intended to prevent overfitting to public test sets [2]. The paper evaluates a broad range of outlier detection methods—classical, deep, and foundation models—across multiple hyperparameter configurations [2]. The authors report empirical findings and practical guidelines alongside individual performance figures [2]. The manuscript was submitted to arXiv on February 10, 2026, and revised twice, with the latest version posted on June 13, 2026 [1]. arXiv, which began operating in 1991, is an open-access repository that hosts preprints across physics, computer science, mathematics, and other fields without peer review [6]. As of late 2024, the platform was receiving roughly 24,000 new articles per month [6]. The MacrOData paper appears under the Computer Science > Machine Learning category and is accompanied by experimental tools offered through arXivLabs, a framework that lets community collaborators build features such as bibliographic explorers and code finders directly on article pages [1][4][5].