detect-anomalies-aiops

Apply machine learning to detect anomalies in operational metrics, correlate alerts, and reduce false positives.

Expected: Time series data loaded with regular intervals, missing values handled, features engineered for ML models.

On failure: If Prometheus connection fails, verify URL and network access, if data gaps exist use forward-fill or interpolation, ensure timestamp column is datetime type, check for memory issues with large date ranges (process in chunks).

使用時間序列分析（隔離森林、Prophet、LSTM）、警報關聯和根本原因分析，對營運指標實施人工智慧驅動的異常檢測。透過智慧識別系統指標、日誌和追蹤中的真正異常來減少警報疲勞。當營運團隊因警報量而不堪重負時、當檢測到超出靜態閾值的複雜多指標異常時、當季節性模式使閾值無效時，或當需要在問題影響用戶之前主動預測問題時，請使用。 來源：pjt222/development-guides。