anomaly-detection

Detect anomalies in time-series data by profiling data, establishing a baseline, and flagging observations that deviate from normal behavior. This process applies univariate or multivariate techniques to identify unusual patterns in metrics and event streams.

Choosing anomaly detection methods involves selecting Tier 1 statistics, Tier 2 time-series, or Tier 3 machine learning techniques based on your data profile. This tiered methodology ensures appropriate univariate or multivariate analysis across domains.

Set alerting thresholds for anomaly detection by establishing a baseline and defining detection limits during operational design. Specifying alerting actions and applying tiered evaluation helps reduce false positives in metric dashboards.

Anomaly detection works on time series, tabular data, event streams, and metric dashboards. It identifies outliers and flags observations that deviate from normal behavior across logs and metrics using univariate or multivariate techniques.

Evaluate whether a detected anomaly is real or an artifact by inspecting results against established baselines and data profiles. The workflow includes evaluating detected anomalies to determine if suspicious data points require further investigation.

Statistical anomaly detection, or Tier 1 methods, may lack effectiveness for complex multivariate data compared to Tier 3 machine learning techniques. Method selection depends on data characteristics, and evaluating tiered results prevents missing unusual patterns.