MinneBOS - Boston's Field Guide to Data Science & Emerging Tech

If you have a data pipeline, you will have data anomalies in it sooner orlater.  If you don't notice them, at best you'll get degraded model performance, and at worst you'll miss important changes in your business environment.

In this talk you will learn how to apply regression and dimensionality reduction techniques to the anomaly detection problem. We'll also discuss investigation and diagnosis of the potential anomaly once it has been detected. Don't be intimidated by all the papers about novel anomaly detection algorithms - idiosyncratic algorithms are not equired, and in fact can sometimes produce warnings that are harder to diagnose than using simpler techniques.

To benefit from this talk, attendees should have a conceptual familiarity with at least one of linear regression, principal component analysis, and chi-squared tests.