Anomaly Detection Process
Using Tangent for anomaly detection is straightforward—here’s how it works:
What You Need:
Time Series/IoT Data: The raw data that reflects the behavior you want to monitor.
Configuration: A setup that defines how Tangent should process the data.
How It Works:
Generate a Normal Behavior Model:
Tangent uses your time series data and configuration to create a model of normal behavior over an extended period. This model captures what is considered typical or expected within the data.
Detect Anomalies:
Once the normal behavior model is established, you can use it to detect anomalies during an out-of-sample period. This involves identifying any deviations from the established normal behavior, whether they are structural (consistent patterns) or sporadic (isolated incidents).
Key Differences from Forecasting:
Static vs. Dynamic:
In anomaly detection, the model for normal behavior remains static for longer time periods—it’s built over time to encompass all possible normal conditions. This helps in spotting deviations effectively.
In contrast, forecasting involves dynamically building and applying models simultaneously to enhance accuracy and reduce model drift.
Tangent offers seven anomaly indicators designed to capture a wide range of anomalies. This ensures that you can detect everything from single-point multivariate deviations to structural patterns that unfold over longer time periods.
Category | Subcategory | Supported by Tangent |
---|---|---|
Anomaly Detection Type | Normal behaviour model with target | ✅ |
Normal Behaviour model without target | ❌ | |
Labeling Strategy | Supervised | ✅ |
Semi-Supervised | ✅ | |
Unsupervised | ❌ | |
Input data type | Continuous data | ✅ |
Categorical data | ❌ | |
Input data dimension | Univariate | ✅ |
Multivariate | ✅ | |
Temporal context | Continuous processes | ✅ |
Event-driven processes | ✅ | |
Batch processes | ❌ |