Meerkat: Anomaly Detection as a Service – Julien Herzen
Julien will present Meerkat, a system built at Swisscom to do real-time anomaly detection on time series. Meerkat uses a combination of machine learning and big data technologies in order to trigger alerts in case of problems in Swisscom network.
Meerkat monitors arbitrary time series and trains statistical models that can be used to spot anomalies from both batch (historical) and streaming (live) data. It is composed of a Python modules for anomaly detection and data ingestion from Druid, as well as Scala modules using Apache Spark for ingesting from Apache Kafka and Apache Hadoop’s HDFS.
Meerkat is currently successfully used at Swisscom to trigger alerts in case of problems with VoIP calls, which represent more than 3 millions phone calls per day.
This is joint work with Khue Vu, who worked on Meerkat for his MSc thesis at EPFL, and the network intelligence team of Swisscom Innovation.