1

I am looking for algorithms on Anomaly detection for time series data. It is uni-variate analysis, considering single parameter (inlet pressure) of air compressor sensor data. The objective is to detect if there is any significant deviation in parameter value in real time.

If any one has worked on similar projects, please share your thoughts.

Ethan
  • 1,633
  • 9
  • 24
  • 39
Sunil M
  • 11
  • 4
  • Can you describe in more or less specific terms the nature of your time series, and of your anomalies? – Laurent Duval Apr 27 '16 at 07:48
  • Semi supervised SVMs are quite efficient on IT security matters. You need a few supervised learning to get a good anomaly detection. See that for an example: http://rslab.disi.unitn.it/papers/R86-TGRS-Jordi.pdf – Robin Apr 27 '16 at 22:22
  • You may have a look at this https://datascience.stackexchange.com/questions/6547/open-source-anomaly-detection-in-python/6549#6549 – Kasra Manshaei Oct 26 '18 at 14:38

1 Answers1

1

Welcome to DS forum. There are many methods available for anomaly detection of uni-variate, low dimension and high dimension data sets.

Supervised algorithms can be used if the past anomalies are recorded with flag variables and there is significant examples of flagged anomalies.

In unsupervised algorithms the key aspect of anomaly is which subset of the larger data set is considered when looking for anomolies i.e., global vs local and point vs contextual

Anomaly Types

Here are some useful links to get you started, anomaly detection in machine generated data and introductory overview of time-series based anomaly detection algorithms.

Hope this helps.

Manohar Swamynathan
  • 696
  • 5
  • 11