irregular time series prediction
On 9/7/05, jfontain at free.fr <jfontain at free.fr> wrote:
Hello. This is my first post, so allow me to introduce myself. But first, I'd like to thank all the authors and contributors to the R software, as I think that it is truly a great and very useful package. I am the author of moodss, a GPL modular monitoring application (http://moodss.sourceforge.net). Moodss collects, archives in a SQL database and displays data from monitored devices, mostly computers, databases and network equipment. My idea is to use the stored data to perform predictions for capacity planning purposes. For example, based on the trafic on a network line for the last 12 months, what is the expected evolution in the next 3 months. Since there is no guarantee that the data samples are regularly spaced in time, I was thinking of using the "its" package for a start. But data samples, most of the time, are roughly regularly spaced. For example, the monitored network device could return data every 10 seconds, but sometimes at 11 seconds or 9 seconds after the last sample. So another idea would be to normalize the data (by interpolation maybe) to make it a regular time series as a first step. All I need from you at this time is to point me in the right direction, maybe suggest resources on the web about this subject applied to R, knowing that I would prefer to use only R base packages if possible. I apologize if this is a trivial question, but last time I studied statistics was more than 20 years ago, so I need a little time to warm up...
The zoo package has support for irregular series and weakly regular
series (classes 'zoo' and 'zooreg') where the latter are series that have
an underlying regularity, e.g. every 10 minutes, but there may be some
missing ones. Also you might want to look at the article on dates
and times in R News 4/1.
library(zoo)
vignette("zoo")