Researchers Release Large Open-Source Dataset for AI Root Cause Analysis

Researchers have unveiled LEMMA-RCA, a comprehensive dataset designed to accelerate progress in root cause analysis (RCA) — the process of identifying why complex systems fail. The resource addresses a long-standing gap: the lack of large-scale, publicly available benchmarks for training and evaluating RCA algorithms.

What the Dataset Contains#

LEMMA-RCA spans multiple domains and data types, featuring fault scenarios drawn from both Information Technology (IT) systems — such as microservices architectures — and Operational Technology (OT) systems including water distribution and treatment infrastructure. The dataset encompasses hundreds of system entities, capturing real-world conditions that machine learning models need to handle.

Testing and Performance#

The team evaluated six baseline methods across different configurations using LEMMA-RCA: comparisons between offline and online models, as well as single and multi-modal approaches. These tests demonstrate the dataset's capacity to support rigorous, standardized evaluation of RCA techniques across varied scenarios.

Why It Matters#

By opening this dataset to researchers, the work enables fairer comparisons between competing RCA methods and encourages development of more reliable diagnostic tools for complex systems. Public, domain-diverse benchmarks like this are essential for advancing techniques that help engineers understand and prevent system failures.

The dataset is publicly available via arXiv, with archival metadata and access instructions provided under a DOI identifier.