Reduction of the Boasting Bias of Linear Experts
If no large design data set is available to design the Multiple classifier system, one typically uses the same data set to design both the expert classifiers and the fusion rule. In that case, the experts form an optimistically biased training data for a fusion rule designer. We consider standard Fisher linear and Euclidean distance classifiers used as experts and the single layer perceptron as a fusion rule. Original bias correction terms of experts’ answers are derived for these two types of expert classifiers under assumptions of high-variate Gaussian distributions. In addition, noise injection as a more universal technique is presented. Experiments with specially designed artificial Gaussian and real-world medical data showed that the theoretical bias correction works well in the case of high-variate artificial data and the noise injection technique is more preferable in the real-world problems.
Unable to display preview. Download preview PDF.
- 2.Ho T.K.: Data complexity analysis for classifier combination. Multiple Classifier Systems. Lecture Notes in Computer Science. Springer Vol. 2096. (2001) 53–67.Google Scholar
- 6.Raudys Š.: Combining the expert networks: a review. In: Sadykhov R.K. (ed.): Proc. of Int. Conference Neural Networks and Artificial Intelligence, ICNNAI’2001. BSU publication, Minsk (2001) 81–91.Google Scholar
- 8.Bedworth M.: High Level Data Fusion. PhD Thesis, Aston Univ, UK (1999).Google Scholar
- 9.Alkoot F.M., Kittler J.: Improving the performance of the product fusion strategy. In: Proc. 15th IAPR International Conference on Pattern Recognition, Barcelona, (2000).Google Scholar
- 10.Raudys Š.: Experts’ Bias in Trainable Fusion Rules. In: IEEE transaction on Pattern Analysis and Machine Intelligence, (2001) (submitted).Google Scholar
- 11.Kahya Y.P., Güler E.C., Sankur B. and Raudys Š.: Hierarchical classification of respiratory sounds. In: Chang H.K. and Zhang U.T. (eds.): Proc. of 20th annual. Int. Conf. of IEEE Engineering in Medicine and Biology and Biology Society, Pts 1-6. Biomedical Engineering towards the year 2000 and beyond. Vol. 20 (2000) 1598–1601.Google Scholar
- 12.Janeliūnas A.: Bias correction of linear classifiers in the classifier combination scheme. In: Sadykhov R.K. (ed.): Proc. of Int. Conference Neural Networks and Artificial Intelligence, ICNNAI’2001. BSU publication, Minsk (2001) 92–99.Google Scholar
- 13.Raudys Š.: On the amount of a priori information in designing the classification algorithm. In: Technical. Cybernetics. Proc. Acad. of Sciences of the USSR. Nauka Moscow, N4 (1972) 168–174 (in Russian).Google Scholar