Read our Technical White Paper
Building a Large Scale Machine Learning-Based Anomaly Detection System
Part 1: Design Principles
Anodot is a real time analytics and automated anomaly detection system that discovers outliers in vast amounts of time series data and turns them into valuable business insights. Using patented machine learning algorithms, Anodot isolates issues and correlates them across multiple parameters in real time, eliminating business insight latency, and supporting rapid business decisions through its uncovered insights.
With its scalable SaaS platform, Anodot provides typically siloed teams – BI, R&D and Devops – with a single, unified system for both business and IT metrics.
Using patented machine learning algorithms, Anodot crunches your time series data to determine its normal range. From then on, it flags any and all anomalies, assigning them a significance score, an automatic assessment of how important an anomaly is based on how “off” the data is, and for how long a period of time. Distinctive features of the automated anomaly detection include:
Sometimes taken on its own, a single anomaly may not be significant. But when grouped together with other anomalies, it becomes extremely important. For example, similar changes across multiple geographies, or while one metric rises another one falls. Noticing these types of correlations manually is nearly impossible, yet Anodot groups and correlates multiple anomalies by design, to bring you the most important insights first, and eliminate alert storms. No manual configuration is necessary to gain these correlated anomalies; Anodot automatically learns from your data and provides you with unique insights.
Anomalies are delivered right to your favorite application with extensive built-in alerting tools. Alerts are automatically grouped so that when there are multiple, related anomalies, a single, unified alert allows you to investigate the phenomenon at once. For example, if there are anomalies for both a publisher and an ad exchange, both of which have alerts configured, Anodot would provide a single combined alert to show the relationship between them. And then, one click on the “investigate” button brings you to the right place in the Anodot interface for further research. Static threshold alerts may also be set, but in most cases are not necessary when tracking anomalies. Alert notifications can be sent via email or webhook (e.g. Slack).
A full suite of functions lets you view your metrics in infinite combinations. Easy, wizard-driven function trees let you create and save composite metrics, and enable quick building of charts and dashboards. Saved composite metrics (e.g. average checkout amount across multiple regions; or the ratio of product views to checkouts for a single product or group of products) are then tracked for anomalies moving forward.
One-click charts can be displayed as lines, columns, area and spline, with stacking options. Every chart enables drill-down into the data for further investigation. Charts can be grouped together into dashboards for easy viewing and handling, giving you quick-access bookmarks for often-viewed metrics.
Whatever type of time series data you have, Anodot can analyze it and determine its anomalies in real time. With simple integrations to standard metrics solutions such as Graphite, StatsD, Coda Hale Metrics and collectd, it’s easy to get started. Whether in adtech, ecommerce , IoT , or any other data-centric business, start streaming your data to Anodot, and begin gaining new insights that will help improve your business.