5
$\begingroup$

I am watching the seminar (around min 20) from CERN about Higgs' Boson discovery.

They quickly go on event classification talking about boosting decision trees. I am not really aware on what they need to classify -I guess they assume the audience knows enough the project in depth. I wonder if someone knows the whole detection and classification process in subject. The problem statement is fundamental for me to understand it.

Update

I am investigating on what features, classes and statistical measures have been used in the physics experiment and then feed to decision trees and multivariate analysis.

$\endgroup$
2
  • 3
    $\begingroup$ I quite like this explanation from phdcomics. Of course, it is very basic. $\endgroup$ – user10525 Jul 5 '12 at 14:52
  • $\begingroup$ @Procrastinator, thanks to have linked it, really enjoyable $\endgroup$ – stackovergio Jul 5 '12 at 15:14
3
$\begingroup$

Procrastinator has given you a story about Higgs Boson, but that really doesn't explain how decision trees and boosting enters into it. I will give you a general description of these classification algorithms. The classification problem involves taking groups that are categorized and a set of data with several variables (called features) that describe the points. The idea is that points from different categories will have very different combinations of values for these features. Given a set of points with known categories and associated features the classification algoithm will assign a catgory based on the value of the features. This is done to identify the category for a new point whose category we do not know.

The original data with the categories known is called training data and is used to construct the classification rule. A decision (or classification tree) is a binary structure shaped like the branches of a tree. There are two branches at each node. The baranches are constucted so that given certain values of the feature you take one branch and for the remaining values you take the other each branch leads to a node where additional branches are formed.

At each node branches are form using one of the features. This is followed down to terminal nodes at which the category is selected. The order of the the features on the tree and the way the values are split down the branches are determined based on the training data in order to correctly classify the training data accurately. Accuracy is determined by applying the rule given by the tree to test data. There are potentially many trees that can do a good job at classifying the points.

In boosting several trees (called an ensemble) are constructed and the object is classified by all the trees. Then a rule such as majority vote is used to make the decision. This is called ensemble averaging and the procedure is called boosting because it has been shown that generally this procedure produces a more accurate classification rule than any individual tree would do on its own.

$\endgroup$
2
  • 2
    $\begingroup$ Reading again my question, I see it is not clear in what I actually ask. I know what decision tree and boosting are; I am investigating in how those techniques have been applied in the physics experiment. In particular, what features and classes and statistical measures have been considered to build up trees. $\endgroup$ – stackovergio Jul 6 '12 at 3:40
  • $\begingroup$ In that case neither the cartoon nor my answer explain that. $\endgroup$ – Michael R. Chernick Jul 6 '12 at 10:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.