11

I have a datasets with information like age, city, age of children, ... and a result (confirm, accept).

To help modelisation of "workflow", I want to create automatically a decision tree based on previous datasets.

I have take a look at http://en.wikipedia.org/wiki/Decision_tree_learning and I know that the problem is clearly not obvious.

I just want to have advice on some algorithm or some libs on this subject what can help me in the contruction of a decision tree based on samples.

Has QUIT--Anony-Mousse
  • 76,138
  • 12
  • 138
  • 194
X-Blaster
  • 1,751
  • 3
  • 15
  • 32
  • "modelisation of workflow" confuses me. What exactly do you want to achieve? Automatically sort new datasets or check whether existing datasets have the right result or what? – Aaron Digulla Oct 13 '09 at 12:38
  • 1
    We have some datasets with legal issues. Some legal case will be rejected only if a personn have an age below 18. This is an obvious example but we want to recreate automatically a decision tree by previous judgement to make a model of the law to edit and refine it after. This is the main reason we don't want a neural network because we can't retrace and proove choices – X-Blaster Oct 14 '09 at 11:46

2 Answers2

14

You should take a look at Weka, a free Java-based supervised learning suite.

After converting your data into Weka's simple text-based .arff format, you should be able to use the GUI or command-line interface to train and test a variety of different classifiers on that data, including:

  • decision trees
  • neural networks
  • rule-based systems
  • support vector machines (SVMs)
  • various types of regression

Experimenting with this interface should allow you to easily try different classifiers and training parameters to determine which ones perform the best on your data.

You can also use an API to integrate Weka into your own source code.

Nate Kohl
  • 35,264
  • 10
  • 43
  • 55
  • I want to implement such a code on the Android platform, but weka is too huge (heap space) for android..is there a way i can run it through some other way – i_raqz Feb 07 '11 at 21:10
  • @raqz: Well, you could build your decision tree on a local machine, then export the tree in the form of java source code and run it on Android. But that won't help if you actually want to do training on Android -- you'll probably have to find an alternative to Weka. Perhaps something that just does decision trees? Googling for "decision tree java" shows at least a few results. – Nate Kohl Feb 09 '11 at 20:32
2

If you want to compare performance of different types of decision trees from Weka, see benchmark results collected in TunedIT.org:

http://tunedit.org/results?d=UCI&a=Weka*tree.

Play with drop-down lists and name patterns of algorithms/datasets to choose which results should be presented.

Marcin
  • 21
  • 1