Real-Time Anomaly Detection for Advanced Manufacturing: Improving on\n Twitter's State of the Art

The detection of anomalies in real time is paramount to maintain performance\nand efficiency across a wide range of applications including web services and\nsmart manufacturing. This paper presents a novel algorithm to detect anomalies\nin streaming time series data via statistical learning. We adapt the\ngeneralised extreme studentised deviate test [1] to streaming data by using a\nsliding window approach. This is made computationally feasible by recursive\nupdates of the Grubbs test statistic [2]. Moreover, a priority queue [3] is\nemployed to reduce memory requirements, where subsets of the required data\nstreaming window are maintained in the algorithm rather than the full list. Our\nmethod is statistically principled. It is suitable for streaming data and it\noutperforms the AnomalyDetection software package, recently released by Twitter\nInc. (Twitter) [4] and used by multiple teams at Twitter as their state of the\nart on a daily basis [5]. The methodology is demonstrated using an example of\nunlabelled data from the Twitter AnomalyDetection GitHub repository and using a\nreal manufacturing example with labelled anomalies.\n