EngineAD: A Real-World Vehicle Engine Anomaly Detection Dataset

View PDF HTML (experimental)

Abstract:The progress of Anomaly Detection (AD) in safety-critical domains, such as transportation, is severely constrained by the lack of large-scale, real-world benchmarks. To address this, we introduce EngineAD, a novel, multivariate dataset comprising high-resolution sensor telemetry collected from a fleet of 25 commercial vehicles over a six-month period. Unlike synthetic datasets, EngineAD features authentic operational data labeled with expert annotations, distinguishing normal states from subtle indicators of incipient engine faults. We preprocess the data into $300$-timestep segments of $8$ principal components and establish an initial benchmark using nine diverse one-class anomaly detection models. Our experiments reveal significant performance variability across the vehicle fleet, underscoring the challenge of cross-vehicle generalization. Furthermore, our findings corroborate recent literature, showing that simple classical methods (e.g., K-Means and One-Class SVM) are often highly competitive with, or superior to, deep learning approaches in this segment-based evaluation. By publicly releasing EngineAD, we aim to provide a realistic, challenging resource for developing robust and field-deployable anomaly detection and anomaly prediction solutions for the automotive industry.

Comments: 12 pages, 2 figures

Subjects:

Machine Learning (cs.LG)

Cite as: arXiv:2603.25955 [cs.LG]

(or arXiv:2603.25955v1 [cs.LG] for this version)

https://doi.org/10.48550/arXiv.2603.25955

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Hadi Hojjati [view email] [v1] Thu, 26 Mar 2026 22:47:45 UTC (417 KB)