Reference for time series analysis

[yzderiv]

Does anyone happen to come upon any good reference on time series analysis using machine learning techniques?

[pwaller]

One technique is called dynamic time warping. Here are a few links. Afraid I haven't used it myself, but found it interesting to read about. Not sure if it is applicable to your specific problem as you didn't specify what kind of problem you seek to solve :)

• https://en.wikipedia.org/wiki/Dynamic_time_warping
• https://www.cs.unm.edu/~mueen/DTW.pdf
• http://www1.cs.ucr.edu/faculty/philip/papers/conferences/asilomar14/asilomar14-dtw.pdf
• http://alumni.cs.ucr.edu/~ratana/RatanamC.pdf

[yzderiv]

Thanks a lot! The problem is to analyze and forecast customer demand in option markets. These are usually similar but slightly different contracts traded by various kinds of customers.

I will have a look at the references.

[betatim]

In a trading setting I have seen people setup the problem like this:

• currently at time n
• want to predict value of time series at n+1
• use k previous days to make a prediction for "today", so split the time series into chunks of k observations.
• build a model that has k inputs to predict the value for today

Straight forward cross-validation doesn't work anymore in this case, in sklearn there was a recent addition called http://scikit-learn.org/stable/modules/generated/sklearn.model_selection.TimeSeriesSplit.html#sklearn.model_selection.TimeSeriesSplit which might help.

[yzderiv]

Thank you very much. I will have a try.