Cross-Industry Anomaly Detection Solutions with R and Shiny
summary
Uses the Twitter-derived R anomaly package (hrbrmstr fork with tibble/snake_case) and a real-time Shiny app to run lightweight time‑series anomaly detection (defaults to piecewise median over two‑week windows and uses one year of history). Built a production Shiny pipeline that flags hardware sensor failures before shipment and surfaces weekly spikes in sports news (also caught a diffbot tagging bug). Method is four times faster than heavier options and proved effective after iterative parameter tuning and business‑rule integration.
AI-written summary · about
chapters (13)
- 0:00Intro
- 0:22Quick Intro
- 1:00What is Anomaly Detection?
- 1:39The R Package
- 2:20Piecewise Median outperforms STL and Quantile Regression B-Spline for Anomaly Detection on Time Series Data
- 3:25Implementing Complex Technical Solutions: What Could Possibly Go Wrong?
- 6:00Real World Example: Life Sciences
- 6:49The Data and Experimentation
- 8:50The Code
- 9:59Combine Client Threshold Rules with Anomaly Detection
- 11:17Real World Example: Sports & Entertainment
- 15:16Final Thoughts
- 17:21Questions?
related talks
18mCross-Industry Anomaly Detection Solutions with R and Shiny
42mDetecting novel anomalies in Twitter
39mMastering real-time anomaly detection with open source tools
29mREAL-TIME ANOMALY DETECTION MADE EASY
45mAnomaly Detection in Time Series
30mBetter data quality with ML & visualization
43mDetecting Anomalies in Streaming Data
26mAnomaly Detection System for the CERN Cloud