0

I have time series data set which contain TimeStamp[hour base] and wind sensor value. I need to find anomalies from this data set.

What are the techniques to find out anomalies ?

How to find anomalies with only these two features ( TimeStamp, sensor-value ) ?

user3666197
  • 1
  • 6
  • 50
  • 92
manju badiger
  • 21
  • 6
  • 2
    please share your research; what did you find and how is this not (completely) solving your problems and challenges? – Chrisvdberge Nov 28 '19 at 07:41
  • 1
    here is an article that might help you get started https://blog.floydhub.com/introduction-to-anomaly-detection-in-python/ – Joran Beasley Nov 28 '19 at 07:43

1 Answers1

0

Very broad question so this will be a generic/broad answer:

To define anomalies you'll need to think and define what you consider normal. Usually we consider two things in terms of (time series) data;

  1. data availability: is the data there that you expect? Usually monitorred by looking at a row count over time (are you inserting more or less data than expected) also counting null values could be used here, but this already leads into the question of data quality:
  2. data quality: are the values in ranges you expect? are they in the type/format you expect, etc. you can use standard deviations/variance/normal distribution to monitor this. Or hard limit and define which values you accept/expect (min, max for instance)
Chrisvdberge
  • 1,824
  • 6
  • 24
  • 46
  • Steps do to detect anomalies: 1. Compute the error term(actual- predicted). 2. Compute the rolling mean and rolling standard deviation(window is a week). 3. Classify data with an error of 1.5,1.75 and 2 standard deviations as limits for low,medium and high anomalies. (5% of data point would be identified anomalies based on this property) Is this right approach ? – manju badiger Nov 28 '19 at 07:50