Anomaly and outlier detection… Neat, advanced stuff in this space that I finally got my hands dirty with in R. The
anomalize package, created by Matt Dancho, helped me identify outliers and smooth out my forecasting model (backtesting also improved). I used
anomalize in conjunction with Facebook’s
prophet for a time-series projection. For R users out there: This package works beautifully with the tidyverse.
I’m continuously impressed by the stuff that Matt puts out including educational content, tutorials, webinars, and his package development. I followed one of his posts on Business Science to learn how to use his anomaly detection package.
FREE Learning Lab for Anomaly Detection:
- If you aren’t aware of Matt Dancho’s Learning Labs, every few weeks his company, Business Science, hosts educational webinars on a variety of topics. His next learning lab will cover the
anomalizepackage to show you how to detect time series anomalies at scale to investigate 80 million rows + 150 groups of web traffic time-series data.
- Webinar Registration: Learning Lab 18
Links to learn