Classification problem

avion

Hello all,

I am a newbie with rapid miner framework and also I have very little experience with machine learning so I would appreciate any help with my learning problem. I have training and test dataset which look like this

a	b	c	d	...	x1	x2	x3
0.3	0.7	20	A	...	0.5	0.8	M
0.4	0.2	10	B	...	0.3	0.9	N
...

All attributes/columns are numerical except d and x3 are nominal.
Problem is to classify test data (attributes x1, x2, x3) based on training dataset. If x1, x2 and x3 would be independent I could create 3 separate programs and learn each parameter independently from another. What learners should I use? Is this even posible?

Please point me towards the solution.

fischer

Hi Avion,

you are asking some rather general questions there. Your first problem is probably to read the data. Use an ExampleSource and the attribute editor to define column types and names. If you want to generate three independent models, you can declare any of x1, x2, and x3 as the label, using a ChangeAttributeRole. You can also iterate automatically using the MultipleLabelIterator. If your three labels are dependent you may want to look at the VectorLinearRegression. However, all of the above may be a bit advanced if you are an absolute newbie.

In general, it is impossible to tell which learner suits your problem best without knowing more about the problem. I would recommend either getting a good data mining book or looking at the training courses:

http://rapid-i.com/content/blogcategory/8/142/lang,en/

A good starting point is also the RapidMiner online tutorial and the sample processes that come with RapidMiner.

Best,
Simon