To check out the power of numeric prediction I have a dataset with currency rates over two month. what I have to do is to find out where will e.g. the currency rate of usd/chf, be in one week. what could be a good approach for this topic?
I'm testing with the MLP Node but it doesn't seem to deliver any result.
Sorry, I must have completely misunderstood your question and the fact the you are trying to predict the currency rate in future. I am not sure, if an MLP is an appropriate solution for this task, I guess, it's hard to handle date/time within the network? Two other things that came to my mind are moving averages and regressions, maybe something to check out in KNIME.
I transformed my workflow and tried it again. My data looks like:
Ccy; Day1; Day2; Day3
USD/CHF; 1.07; 1.06; 1.05
USD/EUR; 1.12; 1.13; 1,12
USD/GBP; 1.23; 1.24; 1.26
What I want are the values for Day 4. I trained a MLP and a Linear Regression Learner. Both delivered results. At the moment the results of the Linear Node are better.
But what I'm a bit confused about is the defining of the target attribute (Day4). Because I have to select e.g. Day3 as target and then I get predicted results from my test data for Day3. So I can only get results if I have already values for my target attribute. Am I right? So there wouldn't be any real prediction because I have to wait until Day4 to train the system, but that wouldn't make any sense.
I have the same problem but with stocks. For example, I have the date,open, close, max, min and volume data for a stock value. I want to have the close price for (date + 1). I´m using linear regression but I think that the prediction is for today and not for date+1.
On the transformed data, you can apply the Math Formula node, since it works on indiviual columns However, the Java Snippet might also be a good choice when you want to refer to rows/column values of previous rows. For an example see this thread.