predicting user actions while using GUI/application

Question

I am using Java (but am open to solutions in other languages as well). I am looking at open source predictive modeling solutions for guessing what GUI/application features a user is interested in (I will have the specific user behavior data on the GUI/application). Instead of just looking at most used actions etc, should I possibly look at incorporating SVM or decision trees? I am looking at weka, mahout and jahmm - is there any other resource I can look at (specifically for GUI behavior - which hopefully returns results fast enough even if accuracy is reduced). Since I am not extremely knowledgeable about this field, please inquire about any information I may have left out to better ascertain a working solution. Thanks!

Answer 1

It's incredibly difficult to say given that we don't know what data you're using (I don't know of existing software to do this, but it may very well exist). With respect to support vector machines, they are binary or one-versus all classifiers, so I don't think they would be applicable here, if I understand your intentions correctly.

If you're unfamiliar with machine learning, Weka may be a good place for you to start. If you have supervised data, then you can feed all of your feature vectors with associated classification data into Weka and use cross-validation to see what type of technique suits you best. Additionally, you can use Weka to see if certain features are more important than others and do manual dimensionality reduction. Or of course, you can use one of Weka's dimensionality reduction techniques, but it may be difficult to decide which one if you don't know the assumptions that they make or how your data is related (this also applies to whatever prediction technique you try/use). Although, if you have enough time, you can just play around and manually just see what works best.