Anomaly Detection Session 6: Data long term drift or change point detection

December 18, 2020
Start Time
9:00 AM PST
End Time
10:00 AM PST


This is part of Machine Learning-Driven Anomaly Detection

What you will learn

For a model built with historical data, model drift or concept drift occurs when there is a significant change in the model behavior with the passage of time due to the nonstationarity of data 

Concept drift can be of the following types with respect to classification problems

  • Change p(y | x) i.e the class boundary changes without change in p(x). It’s called real drift. The trained model dictates p(y|x)
  • Change in p(x) only called virtual drift
  • Change in p(x) and p(y|x)

Drift Algorithms we will cover:

  • Drift detection method (DDM)
  • Adaptive windowing  (ADWIN)
  • Early  drift detection method (EDDM)
  • Exponentially weighted moving average concept drift detection (ECDD)
  • Local drift detection (LLDD)
  • Kullback Leibler (KL) divergence
  • Kolmogorov Smirnov test (KS)
  • Cramer Von Mises test (CVM)
  • And many more.

This is a FREE class!


Can’t make it to live session?  No worries.  Go ahead and register; we will send you the session recording.
See below for past session recordings & notes

Intended Audience

COO, CIO, DevOps, Software Engineers


  • Must have: Development experience
  • Nice to have: Python knowledge

What to Bring

  • Please bring a reasonably modern laptop (Corporate laptops with overly restrictive firewalls may not work well;  Personal laptops are recommended)
  • [nice to have] download our docker image elephantscale/es-training
  • Class Notes here

Session Recording

Class Notes

Git Repo

The python implementation is available in the open source  project avenir in GitHub


Pranab Ghosh

Pranab Ghosh is a Data Science Consultant, He owns several open-source Big Data and Data Science projects using Hadoop, Spark, Storm, Kafka, NoSQL databases, and the related ecosystem.