aiops-infrastructure-anomaly-detection

What problem does it solve?

It helps you detect unusual infrastructure and data-pipeline behavior using statistical baselines and machine learning, reducing alert fatigue from hard-coded thresholds.

Core Features & Use Cases

• Prometheus statistical anomaly rules: build rolling baseline alerts (z-score, MAD/IQR, and seasonal-aware signals) for metrics like CPU/memory/disk/network and pipeline throughput/lag.
• Python anomaly detection models: score multivariate anomalies with robust statistics (z-score/IQR/MAD) and Isolation Forest, plus seasonal forecasting with Prophet.
• Alert threshold auto-tuning: generate warning/critical thresholds from historical distributions to target a desired false-positive rate.
• Data pipeline anomaly signals: identify row-count drops, freshness drift, and throughput anomalies using historical windows and z-score style scoring.
• Use Case: When Kafka consumer lag or container memory behavior drifts from its typical pattern, automatically flag the event and route it with severity based on how far it deviates from the learned baseline.

Quick Start

Use the aiops-infrastructure-anomaly-detection skill to generate Prometheus alert rules and Python detection code for a selected metrics set, then review detected anomalies in a Grafana dashboard.