Options

# Beginners Q: Regarding prediction with time series

Member Posts: 3 Contributor I
edited November 2018 in Help

Hello all,

I'm starting to use RM in order to solve a 'problem' I have. I have a linear programming problem already defined which I'll be running with GAMS, but one of the inputs I have is a parameter which I have to "guess" / predict.

The problem context is about supply chain management: plant -> delivery to distribution center -> delivery to customer -> customer. Or:

(P) .......(DC)........(C)

Anyways, one database I have is the previous demand of the clientes back in 30 periods, and I have to predict the next 10.

The data in that Excel sheets comes like this:

t | demand

Where t={ -29, -28, ...., 0} --> historical

And demand is, well, demand.

My Q is: how can I use RM to predict the next 10 periods? Is there any prefered model to do this?

Sidenote: I used STATA11 in order to try to predict demand, but I am not sure of the predictions I got, since I wasn't too sure at the moment of choosing the ARiMA models.

Anyways, hope anybody can help.

Thanks for any input-

Regards,

Tagged:

## Answers

• Options
Moderator, RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,635 Unicorn

Take a look at the "fit trend" operator, which allows you to use different types of algorithms to generate a trendline which can then be used to generate a prediction.

Searching in this forum for time series related posts will also yield lots of helpful information!

Regards,

Brian T.
Lindon Ventures
Data Science Consulting from Certified RapidMiner Experts
• Options
Member Posts: 3 Contributor I

Alright, I found it with an extension of time series (series extension).

But, how do you use it? I'm totally lost in using RM. If it's possible, how would it be, in a "for dummies" way?

• Options
Member Posts: 3 Contributor I

[[Sorry for double posting]]

I did the following process:

The Excel file has the following  columns: Client / Period / Demand

Where client has value: 1, 1, 1...., 2, 2, 2,...., 10, 10.

Period, goes from 1 to 30.

Demand is for each client in each period.

Ex:

1 / 1 / 25

1 / 2 / 20

Then, in the filter example process, I filtered by client = 1, in order to forecast that client demand.

This is where I get many doubts:

I used the "index series" with the period attribute, in order to tell RM that that is the "time" serie.

After that, I used the predict series operator. BUT, I have to use a process inside.

Which process should I use?

Thanks in advance for any help

Regards,

• Options
Moderator, RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 1,635 Unicorn

There isn't one "correct" modeling algorithm that is appropriate for all datasets.  The nice thing about this operator (and RapidMiner in general) is that you have the opportunity to look at different types of inner learners and compare their performance quite easily.  You should probably first look at your time series data to see what type of shape/pattern you observe.   You can start with a baseline linear regression and see what kind of prediction that generates, and then compare its output to some of the non-linear choices such as gradient boosted trees or neural nets/deep learning.

Brian T.
Lindon Ventures
Data Science Consulting from Certified RapidMiner Experts
Sign In or Register to comment.