Realtime Time Series Anomaly Detection in Production
summary
Realtime pipeline that forecasts with NeuralProphet and scores residuals rather than raw values, using minute-level data with rolling forecasts and occasional multi-day horizons. Uses explicit event windows, compares lookback choices (entire history vs 10‑minute rolling vs partitioned weekdays vs weekends), shows mean z‑score with cutoff ~3 produces many false positives while median-based z-scores (median + MAD) remain robust to anomalies and reduce retraining contamination.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 2:09Talk agenda and time series
- 7:52Neural Prophet framework
- 12:29Real-time anomaly detection
- 24:28Building real-time pipelines
- 28:38Q&A
related talks
41mApplying Foundational Models for Time Series Anomaly Detection
43mDetecting Anomalies in Streaming Data
39mMastering real-time anomaly detection with open source tools
41mNeural Prophet – A powerful AI framework for Time Series Models
29mREAL-TIME ANOMALY DETECTION MADE EASY
45mAnomaly Detection in Time Series
41mReliably forecasting time-series in real-time
18mCross-Industry Anomaly Detection Solutions with R and Shiny