1

I am a newbie to this Time series analysis problem, but want to explore more and learn about it. I have a question in which I would like some experts to help me out. I have a value called as "airflow" which is measured every 5 minutes continously for 24 hours, and I have the data for the last 15 days. I want to predict what would be the airflow for the next 2 days learning the data from the past 15 days.

How can I do it? Which methods should I use?

I have been seeing certain videos to predict trends and seasonality but have not been successful in doing that, any help how to move forward in this problem will be much appreciated.

Richard Hardy
  • 54,375
  • 10
  • 95
  • 219
Sahil
  • 121
  • 2
  • Do you have any prior knowledge on how you expect the airflow to behave? Do you expect it to decrease/increase linearly or quadraticly for example? Or is the data more or less 'random' and are you hoping that a learning algorithm can find a pattern in it anyway? – dimpol Oct 24 '16 at 09:03
  • The data is more or less random, I hope to learn an algorithm that fits the historical data and maybe give me a prediction for the next 2 days. – Sahil Oct 24 '16 at 09:37
  • Well, I don't have much experience in python or in predicting time-series. My main advice would be to first try to find a model that tries to predict the airflow of days 12 and 13 based on the data from days 1 to 11. If you find such a model, you can validate it by feeding in the data from days 3 to 13 and see if it can predict the data from days 14 and 15. Especially since the data seems random, you will need to watch out for overfitting. – dimpol Oct 24 '16 at 09:53
  • I recently answered a similar question where the user had 90 days ( you have 15 ) and he had 24 readings per day ( you have 288 ) . Please see http://stats.stackexchange.com/questions/238797/robust-time-series-regression-for-outlier-detection/239979#239979 for more details . – IrishStat Oct 24 '16 at 10:06

0 Answers0