Predictive analytics
Predictive analytics forecasts the response time of your monitors a few checks ahead. When the forecast crosses a threshold, or when the latest value is unusual for that monitor, Mugnsoft can tell you before the monitor itself changes status.
Predictive analytics is optional. It needs the prediction service (a separate Mugnsoft component) and an Integrator that writes monitor data to InfluxDB. Without them, monitors work exactly as before.
| Property | Value |
|---|---|
| What is forecast | Response time of URL, API, TCP, UDP, Ping, DNS, Database, System and WebSocket monitors |
| Granularity | One model per monitor and per probe |
| Forecast | The next 5 checks by default (up to 50) |
| Predicted status | The worst forecast value compared with the monitor’s minor / major / critical thresholds |
| Anomaly | The latest value is more than 3 standard deviations away from what the model expected |
| Training | Automatic when prediction is enabled, retrained every week by default |
| Alerts | Optional, on the monitor’s “on status change” channels, through the Integrators |
| Impact on monitoring | None: a forecast never changes a monitor’s status, events or SLA |
Where to start
| You want to | Read |
|---|---|
| Set it up now, step by step | Tutorial: your first forecast |
| Understand what is forecast, how models are trained and when alerts are sent | How it works |
| Look up a setting of the Webserver or of the prediction service | Configuration |
| Fix a connection, training or alerting problem | Troubleshooting |