Inject Synthetic Anomalies
The performance of an anomaly detector is defined by how well it detects anomalies of interest. Historical data provides one type of anomalous data, but it may not be comprehensive. Simulation can provide additional anomaly signatures, but at the cost of having to develop the simulation model.
An alternative approach is to create synthetic anomalies that can reasonably represent the sorts of anomalies a system is likely to encounter. Time Series Anomaly Detection for MATLAB® provides a set of anomaly models that you can configure for your system. You can then inject these models into your data and test how well your detector identifies them.
Functions
syntheticAnomaly | Define the parameters of an anomaly model that can be injected into a time series (Since R2026a) |
injectAnomaly | Inject anomalies defined by one or more anomaly models into a univariate time series (Since R2026a) |
Objects
NoiseAnomaly | Synthetic noise anomaly model for validating anomaly detection models (Since R2026a) |
DriftAnomaly | Synthetic drift anomaly model for validating anomaly detection models (Since R2026a) |
BiasAnomaly | Synthetic bias anomaly model for validating anomaly detection models (Since R2026a) |
PointOutliersAnomaly | Synthetic point outliers anomaly model for validating anomaly detection models (Since R2026a) |
Topics
- Detecting Anomalies in Time Series
Examine the general workflow for developing anomaly detectors that detect anomalous subsequences in time series.
- Use Synthetic Anomalies to Help Validate Anomaly Detector Models
This example shows how to generate synthetic anomalies that you can use to support the testing and validation of anomaly detector models.
