Discussion in 'Technical Analysis' started by mickmak, Dec 14, 2010.

  1. mickmak


    Has anyone used Weka?

    I just started to read more about Weka and it seems like a very popular data mining tool with machine learning algorithms built in.

    Maybe I just can't connect the dots here. I am not sure how it applies in the world of trading when prediction is the key.

    One of the example I found online is related to fundamental analysis with Weka. You input Profit/loss, debt, revenue, etc as attributes and some initial classes (buy, sell, hold). But I am completely not sure how it produces the results that says the stock should be a buy/sell/hold.

    How does it know the it is correct to buy(hold/sell) without having attributes of future prices?

    I understand the attributes are there as variables for correlations (i.e. if revenue is x, P&L is y, etc, then buy). I can't get how buy is verified without another price is inputted. To me, you know the buy is correct only after later prices confirm that the buy is correct.

    Maybe I am just confusing more readers here. Any explaination would be helpful. Thank you.

  2. Maybe you're talking about cross-validation? If you want to enter a "later price", leave a price out of the data set, and then use this price for validation.

    Correlation does not always equal likelihood, correct, however in certain cases it does, or comes very very close. In other words, whatever methods are used, the results may be similar.
  3. mickmak


    Yes. That is the site.

    The product itself is not very well documented with use case examples, etc. But I found a ton of paper (mostly academic) online that utilized this product for predictive analysis.

    I am not sure what to make of it since I am struggling to understand the use cases. If you have a better idea, let me know.
  4. ddude


    Do you have a link to that one example?

    Without more information about that example it would be difficult to speculate on what/how the prediction was done.

    Weka is a workbench for solving various data mining problems that utilizes a nice collection of machine learning algorithms.

    You might also take a look at Rapidminer as well as R for similar utility.



    "One of the tricks here is to get away from thinking that programs have to be composed with only a simple text editor." -Alan Kay-