“School of Cognitive Sciences”

Back to Papers Home
Back to Papers of School of Cognitive Sciences

Paper   IPM / Cognitive Sciences / 11351
School of Cognitive Sciences
  Title:   Improving classification performance with focus on the complex areas
  Author(s): 
1.  Seyed Zeinolabedin Moussavi
2.  Kambiz Zarei
3.  Reza Ebrahimpour
  Status:   Published
  Journal: Lecture Notes in Computer Science
  Vol.:  5901
  Year:  2010
  Pages:   612-626
  Supported by:  IPM
  Abstract:
In combining classifiers, effort is made to achieve higher accuracy in comparison with the base classifiers that form the ensemble. In this paper, we make modifications to the conventional decision template, DT, method, so that its classification performance is improved in experiments with Satimage, Image Segmentation and Soybean datasets. In our modified version, DT, an elegant strategy in classifier fusion, is used in the first stage of classification task, and in the second stage, the most misclassified classes are directed to a classifier that is specifically devoted to those classes. To identify the most misclassified classes, the confusion matrix of the output of the decision template stage is considered. Experimental results demonstrate the improved performance of the modified version by a 3

Download TeX format
back to top
scroll left or right