Anomaly detection in Zabbix data using neural networks
DataForge AI
If you have many Zabbix metrics, it can be difficult to create a trigger that ignores expected behavior, but reliably triggers problems in the event of an anomaly.
With DataForge AI, you can automatically detect anomalies in your Zabbix metrics using neural networks.
For this purpose, custom models are trained based on training data based on automatic data exports from your Zabbix instance to better understand the expected behavior of your systems. If the behavior is unexpected, the trained model will detect this as an anomaly.
DataForge features
DataForge documentation
Open documentationHow it works
Operating principle
Using DataForge AI, you can train neural networks for anomaly detection that are trained on your own data to fit your metrics.
The AI model learns the relationships between multiple metrics while taking temporal context (such as the evolution of your metrics over time, or relationships to the current weekday) into account.
When running the model on live data, unexpected anomalies in these metrics can be detected automatically
Our user experience
Workflow with DataForge AI
The workflow of DataForge AI revolves around these steps:
- DataForge exports the requested training data from Zabbix
- DataForge trains an AI model with the user-configured parameters
- Test of the model using a different set of training data, to check if the model works well
- Configuration of the DataForge AI runner, to run the model on the Zabbix data in real time, sending the anomaly metrics back to Zabbix
- Configuration of triggers in Zabbix to trigger problems if an anomaly has been detected
Your benefits
- Can detect complex issues in metrics
- Adapts to changes through optional automatic retraining
- Simple configuration & helpful overview
- Fully automated workflow
- Direct connection to Zabbix
