Decision tree or R forest for timeseries prediction

[sanchit]

How to use decision tree or R forest for timeseries prediction? if data is as simple as sales and date columns

[ellen-wiegand]

Is there a reason you would not want to use the time series tools? They use ETS and Arima. 

https://community.alteryx.com/t5/Interactive-Lessons/tkb-p/interactive-lessons/label-name/Time%20Ser...

[sanchit]

Hi, Yes i can use them but my business strictly want to use forrest and tree or regressions for forcasts

[sanchit]

I am not sure how you use forest and decision tree or regression and then use them for forecasting values for future using tools alteryx has, if possible can you please guide me how to combine both or may be give me sample workflow with any data that has can showcase it.

[apathetichell]

Here's a good article on the subject:

https://www.r-craft.org/r-news/tuning-random-forest-on-time-series-data/

The part about using createTimeSlices from caret is key.

[sanchit]

Hi apathetichell I can read the articles, But can you show me how to do this in alteryx quickly??

[apathetichell]

Um what you are asking about is a complicated process to retool a classification model to a time series based model - there is nothing "quick" about it. A traditional forest model isn't quick - and retooling it for a time series is definitely not quick.

[sanchit]

@apathetichell Sure I understand its not easy, Is there a way you can help me out in this, may be tell me steps i can take from alteryx perspective , Can i use regression then timeseries forecast will that be something easy, if you can even create a sample work flow it would really help full for me.

I appreciate you time to get back to me, I didnt meant to offend you. please help me out here if there is any flow you can create will would really helpful.

[apathetichell · Accepted solution]

You should hardcode everything in R.  Or you can try just the createtimeslices segment and then using that as your test data and feeding that into a score tool after you have use Forest tool for the holdout data. Either way - I'd do substantial hard coding in R.