Robust Parametric Empirical Bayes based anomaly detection for flight safety events

Control Charts using Parametric Empirical Bayes (PEB) based on a Robust Cauchy prior and Poisson likelihood have proven useful for automatically detecting spikes in safety-related flight operations event rates. As part of the monitoring work of the Aviation Safety Information Analysis and Sharing (ASIAS) program, a joint government-aviation industry program, many hundreds of time series are being monitored. Automated detection methods are a necessity, and control charts a logical choice of tool, but basic control charts are problematic. For example, proportions charts produce excessive alerts. The robust Cauchy-Poisson PEB approach using numerical optimization is efficiently producing manageable numbers of relevant alerts while remaining relatively simple to administer and interpret.