Easy Real-Time Anomaly Detection

AI-driven contextual awareness for near zero false positives


Streaming Graph Example HR
Novelty-example

Find anomalous data fast

When you don’t know what pattern to look for in a data stream, you just want to find the exceptions, the occurrences that don’t fit the norm, Novelty is simple, fast and accurate. No extra data wrangling or data labeling is required. It’s completely unsupervised. It builds a detailed behavioral fingerprint of your data environment for powerful contextual awareness. Feed it data, and it will score every event according to how unusual it is in your data set, and will also explain WHY an event is novel.

Reduce false positives

New is not always novel. Novelty uses its context awareness to accurately tell the difference. For example, in the diagram to the right, yellow dots are unique, but sometimes one-of-a-kind events are normal. Only the ones with highest novelty scores at the top are genuine anomalies.

Novelty-visualization
Numerical-vs-categorical-data

Analyze categorical data

The most important data is frequently not numeric—entity and user behavior analytics (EUBA), network address and traffic analytics, physical simulation of a real thing (digital twins) etc. Novelty’s built-in AI analyzes categorical data as is, even if it has high cardinality. There’s no need to delay analysis to transform categorical data into sparse, bloated numeric data. Novelty analyzes the existing data and returns detailed scores for novelty, uniqueness, information content, and probability within seconds.

Get answers 1000 times faster

We benchmarked Novelty against a standard Isolation Forest algorithm—the next most commonly used anomaly detection system—on the same data on the same 8-CPU, 16 GB virtual machines. Isolation Forest managed as many as 500 events per second. With Novelty, you can score over 15,000 events per second on a single moderately-sized instance.

Novelty example bar chart
Novelty-transformation-example

Transform data right in the stream

Several data transformation functions are included in Novelty so you won’t need a separate transformation layer before you can use it. Stream raw data in directly.

Leverage your expert knowledge

You can limit or remove the influence of highly novel observations you know are not important. You can also boost particular factors that you know have a high impact on how novel something might be.

You can:
1. Remove individual or groups of observations.
2. Define a rolling window of observations.
3. Delete entire contexts from the system.

Recent posts

Want to read more news and other posts? Visit the resource center for all things thatDot.

Help Center

Streaming Graph Help Docs

Get the help documentation

Novelty Help Docs

Get the help documentation