Challenge #24: ARIMA Time Series

For those of you following along, thank you, you can find the solution to last week’s challenge (challenge #23) is HERE.

This week’s challenge will use the predictive time series tool called ARIMA. If you don’t have the predictive tools you can find the installer at http://downloads.alteryx.com/downloads.html look for the link to “Predictive tools only”. The predictive tools in Alteryx execute the analytics in an open source application called ‘R’, the advantage of using Alteryx vs. R is that Alteryx provides a straight-forward user interface and eliminates the need to program directly in the R language. If you want to read more about what is happening under the hood, here is a link to the Wiki on ARIMA. https://en.wikipedia.org/wiki/Autoregressive_integrated_moving_average

The use case: A retailer would like to forecast how many units of a particular product will be purchased from their locations based on a historical trend.

The source data contains weekly data for 2012 and 2013 details how many units have been moved. Some of the data, however, is populated with NULL values. For the NULL values, please assign the monthly average. If the monthly average is also NULL, assign the annual average.

Objective: Forecast the number of units that will be sold in the six weeks following the available data.

Have fun!

challenge_24_start_file.yxmd

challenge_24_solution.yxmd

Join

Preparation

Predictive Analysis

Intermediate

Time Series

Transform

Predictive

Expert

Accepted answers

All comments

DemandEngineer

I enjoyed this for the most part as I've never worked on forecasting with time-series data.

I thought I was going crazy or using the tools wrong... My High and Low forecast didn't match... Then I finally gave up and checked the solution and it turns out the solution had the same values. But these are different than the supplied Answer in the start file. Please update the start file.

Does the ARIMA tool assume the data is sorted order already as it doesn't ask for a time field... only the frequency.

challenge_24_start_file.yxmd

SgdPackard810

Quirky little challenge to solve the missing data at the beginning; simple ARIMA and TS Forecast tools thereafter (auto settings, no need to perform any seasonal or non-seasonal differencing to stationarize the data, we were told to use ARIMA so no need for holdout data and comparison against an ETS model).

Had a little bit of trouble with my TS Plot tool, the decomposition plot was unavailable even though two full periods (2 years of 52 weeks) were present in the dataset. Would have liked that feature, used in conjunction with the ACF and PACF plots to help me manually agree to the settings chosen by Alteryx within the ARIMA tool.

Oh well! Fun challenge all the same.

challenge_24_Solution_SP.yxmd

JethroChen

Spoiler

challenge_24_jc.yxmd

phottovy

Another challenge down!

challenge_24_ph_solution.yxmd

JoePC

Solution Attached

24.yxmd

AlexC2

haven't used the Time Series tools much, great challenge to try it out.

Spoiler

alteryxchallenge24a - Copy.yxmd

Per

Not too hard (until all the parameters need tweaking that is)

challenge_24_start_file.yxmd

Tokimatsu

Solved.

Spoiler

challenge_24_toki.yxmd

liamholland

Solution:

Spoiler

Solution.yxmd

LHolmes

Great fun! Love the interactive tour in the ARIMA I output anchor. Would be great for Alteryx to include some general good/bad ranges for the accuracy tests or provide in the documentation - I though the results were more in the "fair" zone.

Spoiler

challenge_24_solution_lmh.yxmd

Qiu

Mostly cheating😀

Spoiler

challenge_24_start_file-Qiu.yxmd

StellaBon

Spoiler

After looking at the TS Plot, I had to go with a user specified ARIMA (1,0,0) (1,0,0) and I believe my measures of error are better than the initial solution (For Example: my MASE is lower than 1) but please correct me if I'm off base. This was good practice, thanks!

challenge_24_ARIMA.yxmd

AkimasaKajitani

My solution.

Spoiler

Good challenge for starting to use ARIMA tool.

challenge_24_start_file_AK.yxmd

greggrimmer1

Spoiler

challenge_24_completed.yxmd

A2H

Spoiler

Capture d’écran, le 2020-10-26 à 22.27.36.png

challenge_24_A2H.yxmd

unknown

Solved 🙂

challenge_24_start_file.yxmd

StefanvanWyk

Great, Alteryx did make the prediction very easy to do.

Challenge_24_Stefan van Wyk.yxmd

OwenBData

Good challenge - slightly different numbers (+/- 1) on some values. 🙂

Spoiler

Challenge 24.yxmd

OnionKnight

Good to know about TS tools.

please somebody tell me how to use "TS compare tool" between ARIMA and ETS.

challenge_24_start_file.yxmd

TraceyD

Never even heard of ARIMA before!

TraceyD024.yxmd

rayJ

Same process, first time using ARIMA in Alteryx instead of R. Much faster and most components are here as well.

challenge_24_ray_solution.yxmd

aiahwieder

As nearly as I can tell, this solution should work . . . however, my monthly and annual averages seem to be imperceptibly off, and the resulting forecast refuses to match . . .

challenge_24_solution.yxmd

mceleavey

A nice little intro to the Time Series tools. People are scared of these tools but you really don't need to be.

You should be scared of people who use Alteryx in vertical mode.

Savages.

Spoiler

I first of all used two sum tools to group by the aggregations required, first Year and month, then by Year. To these I applied the average of the units.
Once I had smooshed these back together, I used a formula to impute these averages, first monthly, and then annually for those where no monthly average could be calculated.
Then I simply fed the data into the ARIMA tool and then into the TS Forecast tool to get the calculations.