"TIme series

Question

Hi,

I am also interested in predicting time-series values using the windowing process. 
I am however, a bit stuck at the pre processing stage.

My data looks like

Month  Profit 
1	36
2	204
3	212
4	224
5	196
6	196
7	202
8	189
9	180
10	192
11	198
12	154

I am going to predict the following months from this!. 
what would the best preprocessing algorithm or process be to use. if you have already worked examples please be kind enough to show me.

secondly, 
i am interested in applying this derived data to a cost sensitive learner later on. 
do Rapid miner provide this !!! ( in this case i need to use my own cost matrix obtained )

thanks
Priya

Deon

land · Answer

Hi,
sorry, but I'm not familiar with the WEKA learner and its parameters. So you have to figure out yourself, where to enter the cost matrix.
And I cannot test your process, since I dont have your data. If you want me to debug it, please use an example Set generator, so that I simply can copy it into rapid miner.

Greetings,
  Sebastian

svpriyan · Answer

Thanks for the reply i was trying the same and i got stuck with this when i try to use the i have doubts: Since i am having numeric values in my time series (as given above), DO i need to pre process it. also i could not get clear idea where to define my cost matrix. Your help is greatly appreciated!!!

land · Answer

Hi,
you will need to window your data set. Take a look at the multivariateSeries2WindowExamples operator. I guess, the series_representation is encode_series_by_examples in your case. After this you should have several windows, containing the next month as label and a fixed number of the n last months. This can be used with any learner to predict the next month's outcome.
To your second question: Yes, you can use the metaCost operator for cost sensitive learning.

Greetings,
  Sebastian