cancel
Showing results for 
Search instead for 
Did you mean: 

Alteryx Knowledge Base

Definitive answers from Designer experts.
Announcement | Looking to expand your Alteryx skillset?! Check out the latest set of interactive lessons in Alteryx Academy: Creating Analytic Apps

Performing Time Series Forecasting in Alteryx Designer

Alteryx
Alteryx

Time series forecasting is using a model to predict future values based on previously observed values. In a time series forecast, the prediction is based on history and we are assuming the future will resemble the past. We project current trends using existing data.

 

The following is a high-level description of the two pre-packaged time series methodologies our predictive tools support:

 

ARIMA  is the Autoregressive Integrated Moving Average and is the most commonly used forecasting approach

 

ETS model uses exponential smoothing method and is a commonly used forecasting approach based on a weighted average of past observations with the weights declining in size for more distant past values. In short, all past values are factored in the forecast although with decreasing importance as the values are further back in time.

 

 

Additionally the video walks you through how to confirm the completeness of your data using the Field Summary tool and shows you how to use TS Compare the two analyze the error levels of the forecasting methods.

 


Time Series is comprised of a variety of tools within Alteryx, which are part of the standard Alteryx Designer License. Alteryx customers use predictive analytics to identify patterns found in historical and transactional data to identify risks as well as opportunities. Alteryx Predictive analytic tools are built on Open source R. Alteryx users are not required to know R to execute predictive models because all of the models in Alteryx are packaged into easy-to-use macro tools that only require configuration. All predictive tools are macros, and therefore not a “black box”. Macros provide the user with the flexibility to open all models and dissect the logic, as well as see and modify the R-script(s) being executed.

 

This video provides a brief tutorial of using Times Series tools on historical single family home sales and includes an overview on how to configure the following tools:
Field Summary, ARIMA, ETS, TS Compare, and TS Forecast

Comments
Alteryx Partner

The following article actually provides a better modelfrom the same data used in the Alteryx demo...

http://www.autobox.com/cms/index.php/blog/entry/alteryx-forecast-package

 

  • Is there a test in Alteryx to figure out this kind of major change in the model so that advises us to strip the very beginning of data?
  • Alternatively by having a MARS regression model to the data it self or a moving variance time-series can we understand the point of that discrepancy? What do you think?

Best

Atom

Hey everyone,

 

I use Trial version of Alteryx Designer (11.0)

Does it come with time series tools, I don't find it, if yes, how do I configure it.

 

 

Thanks..

Alteryx Partner

have you installed the predictive capabilities? It's asked at the beginning of the installation

if you may want to install R as well...

 

Check out here; http://downloads.alteryx.com/predictive.html

 

By the way I recommend you to download 11.5...

If you have a period in your data where there was no activity, do you have to feed a row with a value of 0 or does the tool account for this ?

Alteryx Partner

You have to decide on what to do with missing data,

then you can fix it with a formula tool...

 

Most common approach is;

"last observation carried forward, You both preserve the last known value and prevent any look-ahead bias"

 

Other options are imputing;

 

  • mean (Summarize tool)
  • median (Summarize tool)
  • mode (Summarize tool)

  • or replacing null with random observations from timely previous observations (random selection tool used on date filtered data)

  • you can also do a linear interpolation (linear regression on date filtered data and score tool)
  • even a spline interpolation (MARS regression on date filtered data and score tool)
  • you can fill it in with an ARIMA model fitted to observations prior to the missing sections on an iterative manner... (ARIMA toolon date filtered data and forecast tool)

Best

 

In my case it is not "missing" data just in that period there was no activity so i would not want to impute as you suggest.

 

My query is more that in these cases do i need to feed the tool a value for 0 for period that there was no activity or does it automatically assume this is the case?