0

For example, here is a traffic graph from google analytics:

enter image description here

You can see that first there is a spike, and then a long plateau time, and then increase to a slightly high stage (but more volatile), and then drop down to 0.

My question: is it possible to detect/describe them when they occur?

Specifically, to automatically tell me:

  • Is there any stable trend recently?
  • Is there any spike/drop down recently?
  • Is there any upward/doward trend recently?

For example, at 2018 Aug, it tells me there is a spike, and at 2018 Oct, tells me there is new stable trend, and a Feb 2019 tells me that it might enter a higher-level with more volatility.

Reasons for doing this: we might have many time series to watch, so we can automate some simple/obvious tasks.

How can I do it? is there any related Python/R packages? (maybe like https://github.com/pandas-profiling/pandas-profiling )

I think it's possible.

  • We may need to handcode what is stable/spike/dropdown, specify the significant level. and determine the window size.
  • A naive approach can be using predictive time series packages like Facebook Prophet, and send alert when time series is not aligned with the prediction.
cqcn1991
  • 1,145
  • 1
  • 10
  • 16
  • 1
    Look at the differences between consecutive values. – Demetri Pananos Apr 30 '19 at 00:25
  • 1
    if you post your data in a csv file i will try and help you https://stats.stackexchange.com/questions/380599/is-it-possible-to-automate-time-series-forecasting/380634#380634 is where you should start – IrishStat Apr 30 '19 at 02:22
  • @DemetriPananos Great insight! and how should I determine window size? and is there any related doc/paper/blog? and how can I detect trend/seasonal effect? – cqcn1991 May 04 '19 at 22:53
  • @cqcn1991 I can't offer a one size fits all solution, but you should be able to detect periodic behavior via FFT. – Demetri Pananos May 05 '19 at 04:12
  • @DemetriPananos you mean something like seasonal decomposition, like https://stackoverflow.com/a/28284962/1794744 ? – cqcn1991 May 16 '19 at 06:46

0 Answers0