Subscribe to DSC Newsletter

Anomaly detection in Time Series Data - Help Required

Dear Group Members,

I am looking for algorithms on Anomaly detection in time series data. I am working on Air compressor sensor data. If any one has worked on similar projects, please share your thoughts.

Views: 3972

Reply to This

Replies to This Discussion

Hi, 

I have found this Blog with several tools and examples about Anomaly study cases:

https://anomaly.io/blog/

Hope it helps

Carlos

Hello

I came across IntelliPredikt who have developed Time Series based Anomaly detection using HTM algorithm for streaming Starter Motor sensor data.  If you are interested to learn more please reach them on [email protected]

             Thanks Ramesh for the suggestion..:)

bk ramesh said:

Hello

I came across IntelliPredikt who have developed Time Series based Anomaly detection using HTM algorithm for streaming Starter Motor sensor data.  If you are interested to learn more please reach them on [email protected]

     Thanks Carlos for your suggestion...:)

Carlos Lizárraga Celaya said:

Hi, 

I have found this Blog with several tools and examples about Anomaly study cases:

https://anomaly.io/blog/

Hope it helps

Carlos

Sunil,  I've used Kolmogorov-Smirnov test (KS for short) before for outlier detection which is good.  Break up your time-series data into a sliding windows with an option of overlap.  The input is now a matrix of mxn, ie,  m  rows and  n columns.  m is the number of windows and n is the number of time-period per window.  The KS will tell you which window/s  is the outlier, it may be multiple windows or just a single or even no outlier found. What language you develop in? I mainly work in Matlab & Java.  There are tons of others for use in outlier/anomaly detection for time-series, so if you work in matlab or java then I can show you more & where to get those packages from each authors' sites.

Look at the Netflix Surus/RAD package.

There are various packages for anomaly detection, but you can also start with simple things such as -

  • How the data looks. Use scatter plot to see the shape of data.
  • Look at variance, i.e. volatility in business lingo.
  • Creating a TS model .. may be use SMA to start with. Then move on to check the followings - CFE, MAD, MSE and MAPE -  these parameters give a good insight into data behaviors.

I am spreading this in my network, will come back when I get some info for you. Cheers.

i wrote a whole thesis on it using LSTMs http://mars.gmu.edu/handle/1920/10250 . . you may not want to do that so i have a literature review in it too to guide you.

RSS

Videos

  • Add Videos
  • View All

Follow Us

© 2018   Data Science Central ®   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service