locked
Nonlinear correllation (autocorrelation) RRS feed

  • Question

  • HI, I need someone pro answer me how to

    find out a correlation (nonlinear using NN or DNN in Azure) between two time series.

    I want to find hidden connections between 2 time series. For example, temperature and ice cream price/

    how to do that in NN Azure?

    Must I predict this time series with NN (and how to do that also) and only then try to find nonlinear correlation or connection between 2 diffrent time series (Without forecasted) or I can simply compare without any prediction?

    PLz help to implement, step by step, What I must to do to find  nonlinear connections between two or more (if possible) time series?

    • Changed type neerajkh_MSFT Sunday, March 29, 2015 1:29 AM it is a question
    Saturday, March 28, 2015 5:50 AM

Answers

  • Can't you just do cross-correlation between time-series - R has a function ccf that computes cross-correlation.

    However, I would suggest be very clear in your objective on why do you need a correlation.

    If you are trying to use the two time series to create a new prediction function to predict price of ice cream given the temperature, then just remove the time dimension and create a scatter plot between temperature and price.

    May be add a CPI, population, location as additional features for regression to account for economic changes over the time.

    Saturday, March 28, 2015 6:15 PM

All replies

  • Can't you just do cross-correlation between time-series - R has a function ccf that computes cross-correlation.

    However, I would suggest be very clear in your objective on why do you need a correlation.

    If you are trying to use the two time series to create a new prediction function to predict price of ice cream given the temperature, then just remove the time dimension and create a scatter plot between temperature and price.

    May be add a CPI, population, location as additional features for regression to account for economic changes over the time.

    Saturday, March 28, 2015 6:15 PM
  • Thanks for an Idea! 
    Saturday, March 28, 2015 9:33 PM