During my study of the future impact of the covariation among vertex descriptors and follow the evolution of the graph, I extracted patterns of my attributed graph as time series. The classical decomposition of additive time series helped me to see the seasonality, trend and remander. So my problem is that when I extended the duration of my time series I lost the seasonality and trend; does anyone know why? Is it normal to transform a linear regression model into time series ? I would like to have different suggestion to improuve my analysis.
.
before moving to the (rather complex) seasonal arima as suggested by Mahmoud above, you could first review the (much simpler, but still a very good workhorse) Holt Winters exponential smoothing so as to get a good understanding of what is going on when one talks about trend, seasonality and so on :
http://forecasters.org/pdfs/foresight/free/Issue19_goodwin.pdf
.
Hi Uwizeye, How did you extended the duration of your time series? Did you use new real data points or just fitted values?
In fact I transform the sequence of covariation among vertex descriptions obtain in data mining and with R I plotted it as a time series so when I tried to use classical decomposition in order to see everything which affect my time series I didn't find seasonality I don't know if it is the right way to do so
So you just differenced your time series once and that will certainly remove trend from the series and sometimes seasonality too. If you are interested to estimate seasonality then that was not the right approach to analyze your data and you have to fit the seasonal ARIMA model to your data.
.
before moving to the (rather complex) seasonal arima as suggested by Mahmoud above, you could first review the (much simpler, but still a very good workhorse) Holt Winters exponential smoothing so as to get a good understanding of what is going on when one talks about trend, seasonality and so on :
http://forecasters.org/pdfs/foresight/free/Issue19_goodwin.pdf
.
Dear Uwizeye
Please try to read from the followings:
Mining Graph Topological Patterns: Finding Covariations among Vertex Descriptors
Issue No.09 - Sept. (2013 vol.25)
pp: 2090-2104
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/TKDE.2012.154
If you are unable to downlaod it, let me know.
With best regards
http://www.computer.org/csdl/trans/tk/2013/09/ttk2013092090-abs.html
- Badges
- Science topic
- Similar topics
- Computer Science
- Data Mining