1
$\begingroup$

my friend is a chemist and his problem is to predict the level of ozone concentration in a single site. We have the data for the last 12 years.

We want to predict the concentration for the coming years (as much as possible).

I know this data set is small, so my question is is this possible? And how much of data is required if not?

What is the tool/method to use?

Here is a plot of my data:

alt text

any suggestions/ help?

Improve this question
$\endgroup$
2
  • 2
    $\begingroup$ Your question lacks some critical details -- first of all, what should be the protocol of this forecast, i.e. you want to predict the ozone level for next day every day? Predict the mean level in 2020 from 2004-2006? Next, is the ozone levels the only data you have? How it is discretized, one measurement per week/day/hour? Maybe you can pull some other relevant data from other sources? What can be relevant from chemical point of view? $\endgroup$
    – user88
    Commented Nov 23, 2010 at 20:31
  • $\begingroup$ thanks for your effort mbq Yes, i want the mean level in 2006- 2007 or more if that is possible.Regarding data, i guess that is the only data that i got as of now,the measurement is once in a month because the data is available in monthly intervals.if more relevant info is needed then i will try to inquire them after two days.Do you mean the focusing can't be done without other data/info. but you mean the focusing can't be done without other data/info. $\endgroup$
    – Shankarooni
    Commented Nov 25, 2010 at 22:43

2 Answers 2

Reset to default
5
$\begingroup$

Try the forecasting package for r. Specifically, the auto.arima() and ets() functions will model the seasonality and trend in the ozone data and allow you to make monthly predictions for future.

Improve this answer
$\endgroup$
3
$\begingroup$

Since all you have is the data for the series you are trying to predict , the approach should be to construct a Robust Arima Model. A Robust Arima Model can reflect not only auto-projective structure (arima component) but changes in Levels and or Trends over time. The parameters of this model should be proven to not have changed over time and this is also true for the variance of the errors. There may be a change in any required "seasonal Dummies" to reflect a deterministic component as compared to a memory component. If you load your data up on the web and share it with the list , perhaps you can get understanding of the analytics and the analytical tools required. Similar work as been done in forecasting a highly seasonal monthly series entitled "air line passengers". You might google "Airline Passenger series BOX-JENKINS" to get some clues as to how a similar series was handled and severely mishandled by NOT CORRECTLY identifying omitted deterministic structure.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.