Extracting characteristics from time series data

[Richard_Steele]

hi

I have a random set of time series data that is calculated after applying an algorithm to a main random time serie data, and really need to extract all the possible characteristics from the set. The goal is to measure those characteristics and perform some statistical graphs based on those measurements ad try to forecast what are the possible next behaviours of the main set.

At the moment I have observed the next characteristics:

1. Trend linearity and curvature
2. Size of peak and trough
3. Time that the value is over 80 or under 20
4. etc...

I think there are much more characteristics that can be observed. The problem is I cannot find a list with characteristics.

 

[Hesch]

try to forecast what are the possible next behaviours of the main set.

Reference https://www.physicsforums.com/threads/extracting-characteristics-from-time-series-data.871918/

Suggestions:

- Fourier transform

- Artificial Intelligence

 

[FactChecker]

There is a very general model for time series called Auto Regressive Integrated Moving Average (ARIMA). It looks for dependencies between the current and prior values. There is a Box-Jenkins technique for modeling ARIMA processes and the tools are available in the R statistical package. See https://en.wikipedia.org/wiki/Autoregressive_integrated_moving_average and http://www.statoek.wiso.uni-goettingen.de/veranstaltungen/zeitreihen/sommer03/ts_r_intro.pdf

 

[Richard_Steele]

Thank you very much guys, I have tried the techniques you mentioned and it worked very good, specially the ARIMA.