Untrained Method for Ensemble Pruning and Weighted Combination

Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 8866)


The combined classification is an important area of machine learning and there are a plethora of approaches methods for constructing efficient ensembles. The most popular approaches work on the basis of voting aggregation, where the final decision of a compound classifier is a combination of discrete individual classifiers’ outputs, i.e., class labels. At the same time, some of the classifiers in the committee do not contribute much to the collective decision and should be discarded. This paper discusses how to design an effective ensemble pruning and combination rule, based on continuous classifier outputs, i.e., support functions. As in many real-life problems we do not have an abundance of training objects, therefore we express our interest in aggregation methods which do not required training. We concentrate on the field of weighted aggregation, with weights depending on classifier and class label. We propose a new untrained method for simultaneous ensemble pruning and weighted combination of support functions with the use of a Gaussian function to assign mentioned above weights. The experimental analysis carried out on the set of benchmark datasets and backed up with a statistical analysis, prove the usefulness of the proposed method, especially when the number of class labels is high.


Machine learning Classifier ensemble Classifier combination Ensemble pruning Weighted fusion Untrained aggregation 


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    Alpaydin, E.: Combined 5 x 2 cv f test for comparing supervised classification learning algorithms. Neural Computation 11(8), 1885–1892 (1999)CrossRefGoogle Scholar
  2. 2.
    Biggio, B., Fumera, G., Roli, F.: Bayesian Analysis of Linear Combiners. In: Haindl, M., Kittler, J., Roli, F. (eds.) MCS 2007. LNCS, vol. 4472, pp. 292–301. Springer, Heidelberg (2007)CrossRefGoogle Scholar
  3. 3.
    Demšar, J.: Statistical comparisons of classifiers over multiple data sets. J. Mach. Learn. Res. 7, 1–30 (2006)zbMATHMathSciNetGoogle Scholar
  4. 4.
    Frank, A., Asuncion, A.: UCI machine learning repository (2010),
  5. 5.
    García, S., Fernández, A., Luengo, J., Herrera, F.: Advanced nonparametric tests for multiple comparisons in the design of experiments in computational intelligence and data mining: Experimental analysis of power. Inf. Sci. 180(10), 2044–2064 (2010)CrossRefGoogle Scholar
  6. 6.
    Jackowski, K., Krawczyk, B., Woźniak, M.: Improved adaptive splitting and selection: the hybrid training method of a classifier based on a feature space partitioning. Int. J. Neural Syst. 24(3) (2014).Google Scholar
  7. 7.
    Rao, N.S.V.: A Generic Sensor Fusion Problem: Classification and Function Estimation. In: Roli, F., Kittler, J., Windeatt, T. (eds.) MCS 2004. LNCS, vol. 3077, pp. 16–30. Springer, Heidelberg (2004)CrossRefGoogle Scholar
  8. 8.
    Rokach, L., Maimon, O.: Feature set decomposition for decision trees. Intell. Data Anal. 9(2), 131–158 (2005)Google Scholar
  9. 9.
    Wozniak, M.: Experiments on linear combiners. In: Pietka, E., Kawa, J. (eds.) Information Technologies in Biomedicine. AISC, vol. 47, pp. 445–452. Springer, Berlin / Heidelberg (2008)CrossRefGoogle Scholar
  10. 10.
    Woźniak, M., Graña, M., Corchado, E.: A survey of multiple classifier systems as hybrid systems. Information Fusion 16, 3–17 (2014)CrossRefGoogle Scholar
  11. 11.
    Wozniak, M., Zmyslony, M.: Designing combining classifier with trained fuser - analytical and experimental evaluation. Neural Network World 20(7), 925–934 (2010)Google Scholar

Copyright information

© Springer International Publishing Switzerland 2014

Authors and Affiliations

  1. 1.Department of Systems and Computer NetworksWrocław University of TechnologyWrocławPoland

Personalised recommendations