Geek asks and answers

Weka AdaBoost does not improve results

In my bachelor's thesis, I should use AdaBoostM1 with the MultinomialNaiveBayes classifier on the problem of text classification. The problem is that in most cases, M1 is worse or equal to MultinomialNaiveBayes without promotion.

I am using the following code:

AdaBoostM1 m1 = new AdaBoostM1();
m1.setClassifier(new NaiveBayesMultinomial());
m1.buildClassifier(training);

So, I do not understand how AdaBoost will not be able to improve the results? Unfortunately, I could not find anything about this on the Internet, as most people seem to be very happy with AdaBoost.

+3
source share
3 answers

AdaBoost - //2- , , 1/2 . AdaBoostM1 M-, - , 1/2, , 1/M. / , , , AdaBoost, . , Kappa Informedness (AdaBook).

, .. , , , , , . , M = 100 1%, 50% AdaBoostM1.

( , ), - . , , .

( MATLAB xls ) : http://david.wardpowers.info/BM - : http://aclweb.org/anthology-new/E/E12/E12-1035.pdf

Adaboost .

+5

. , bias , . , . , , AdaBoost + Naive Bayes Naive Bayes, , Naive Bayes ( AdaBoost - ).

+4

More articles:


All Articles