Association Rules to predict time series

[Helder Borges]

Hi.

I have 7 time series, 1 with historical sales and 6 with historical variables.

I want to make prediction of future sales.

I already used Time Series algorithms.

Now, i would like to use an Association Rules algorithm that could let me predict future sales, so i could compare both.

Since AR works like if x then y, i wanted to know if its possible to use AR for this tasks with the idea "if past sales then future sales".

Can someone help me, telling if it is possible, and if yes, giving me some hints?

Best Regards,
Helder Borges

[phil123]

Hi,

You could use association rules. But association rules do not consider the sequential order. For example, if you have a rule X ==> Y with a confidence of 60 %, it just means that if X appears then Y is likely to also be there with a 60 % confidence. But it could be before, at the same time or after, because association rules do not consider time.

If you want rules that consider time, you could look at sequential rules. A sequential rules is a little bit like an association rule. But it consider the sequential ordering.

If you are interested in this, I have written a paper about using sequential to predict the next webpages that someone will visit. You can read it here:

[url]http://www.philippe-fournier-viger.com/sequential_rules_prediction_2012.pdf[/url]

Fournier-Viger, P. Gueniche, T., Tseng, V.S. (2012). Using Partially-Ordered Sequential Rules to Generate More Accurate Sequence Prediction. Proc. 8th International Conference on Advanced Data Mining and Applications (ADMA 2012), Springer LNAI 7713, pp.431-442.

Also, if you want some association rule or sequential rule mining algorithm implementations in Java to use in your software, you can get source code here:

[url]http://www.philippe-fournier-viger.com/spmf/[/url]

[Helder Borges]

Hi.

Thank you.

I will take a look since i´m reading a lot of papers right now.

I will need to find a way of using association rules. Otherwise i will need to change my master thesis plan :).

If you(or someone reading this post) know something that could help me with information about how to use association rules to predict future sales (profit), i would appreciate a lot.

Best Regards,

Helder Borges

[phil123]

I see.

Then maybe you could have a look at CBA (classification by association rules) and similar algorithms CPAR, CMAR....

The idea is to use association rules to perform a classification task.

I think that you could see your task as a classification task where the goal is to find what is the future amount of profit (the class) from different ranges of profits.

Best,

Philippe

[Helder Borges]

Is that that i want.
I´m rely grateful.
Thank's for the tip :)

Best regards

Helder

[phil123]

You are welcome.

Good luck!