ECoLAD: A Deployment-Oriented Evaluation Protocol for Automotive Time-Series Anomaly Detection

The newly introduced evaluation protocol, known as ECoLAD (Efficiency Compute Ladder for Anomaly Detection), aims to bridge the divide between benchmarking for workstation-class systems and the constraints of real-world in-vehicle monitoring. Typically, time-series anomaly detectors are assessed on unconstrained hardware; however, automotive applications necessitate consistent latency and reliable performance with limited CPU parallelism. ECoLAD employs a monotone compute-reduction ladder that spans various detector families, utilizing integer-only scaling rules and fixed CPU thread limits, while meticulously logging all configuration alterations. This protocol evaluates throughput-constrained behavior by adjusting target scoring rates and measuring coverage—the proportion of detections that satisfy latency requirements. The empirical analysis incorporates proprietary automotive telemetry, with an anomaly rate of around 0.022, alongside relevant public benchmarks. ECoLAD seeks to identify methods that remain viable under deployment-specific constraints, moving past accuracy-focused rankings.