Predictions

NOBA's Predictions engine forecasts future resource usage using time-series analysis. It helps you anticipate capacity problems before they become incidents.

Capacity Prediction

Navigate to Predictions to see forecasts for all tracked metrics.

Supported metrics:

Metric	Description
cpu_percent	CPU usage %
mem_percent	RAM usage %
disk_percent	Disk usage % (per mount)
net_rx_bytes	Network receive rate
net_tx_bytes	Network transmit rate

Forecast Method

NOBA uses seasonal decomposition (STL) combined with an exponential smoothing model to produce forecasts:

1. Trend — long-term direction (e.g. disk filling over weeks).
2. Seasonality — repeating patterns (e.g. daily CPU spikes at backup time).
3. Residual — unexplained variation used to estimate confidence intervals.

The model is retrained nightly using the last 30 days of history (configurable).

Confidence Intervals

Each forecast shows three bands:

Band	Meaning
Median	Most likely value
80% CI	80% of outcomes expected to fall within this range
95% CI	95% of outcomes expected to fall within this range

Wider confidence intervals indicate more volatile or less predictable metrics.

Predicted Breach Date

For metrics with a clear upward trend (e.g. disk usage), NOBA calculates the predicted date to threshold breach:

"At current growth rate, /mnt/data will reach 90% in approximately 18 days (95% CI: 12–27 days)."

A warning alert is raised when the predicted breach date is within the configured horizon (default: 14 days).

Configure the horizon and thresholds in Settings → Predictions.

Per-Service Health Scoring

Navigate to Predictions → Health Scores to see a composite health score for each monitored service.

The health score (0–100) combines:

• Recent uptime (from Monitoring)
• Response time trend (from Monitoring)
• Resource usage trend (from Predictions)
• Security finding severity (from Security)

Health scores are updated every hour.

Prediction Alerts

Configure prediction alerts in Automations → Alert Rules using prediction metrics:

predicted_disk_days_remaining < 14
predicted_mem_percent_7d > 90

These alert before a breach occurs, giving you time to act.

History Required

Accurate predictions require at least 3 days of history. The forecast confidence improves with more data. New installations show "Insufficient data" until enough history is collected.