Artificial Immune Recognition System Based Classifier Ensemble on the Different Feature Subsets for Detecting the Cardiac Disorders from SPECT Images
Combining outputs of multiple classifiers is one of most important techniques for improving classification accuracy. In this paper, we present a new classifier ensemble based on artificial immune recognition system (AIRS) classifier and independent component analysis (ICA) for detecting the cardiac disorders from SPECT images. Firstly, the dimension of SPECT (Single Photon Emission Computed Tomography) images dataset, which has 22 binary features, was reduced to 3, 4, and 5 features using FastICA algorithm. Three different feature subsets were obtained in this way. Secondly, the obtained feature subsets were classified by AIRS classifier and then stored the outputs obtained from AIRS classifier into the result matrix. The exact result that denote whether subject has cardiac disorder or not was obtained by averaging the outputs obtained from AIRS classifier into the result matrix. While only AIRS classifier obtained 84.96% classification accuracy with 50-50% train-test split for diagnosing the cardiac disorder from SPECT images, classifier ensemble based on AIRS and ICA fusion obtained 97.74% classification accuracy on the same conditions. The accuracy of AIRS classifier utilizing the reduced feature subsets was higher than those exploiting all the original features. These results show that the proposed ensemble method is very promising in diagnosis of the cardiac disorder from SPECT images.
KeywordsSingle Photon Emission Compute Tomography Classification Accuracy Independent Component Analysis Feature Subset Independent Component Analysis
Unable to display preview. Download preview PDF.
- 3.Kurgan, L.A., Cios, K.J., Tadeusiewicz, R., Ogiela, M., Doodenday, L.S.: Knowledge discovery approach to automated cardiac SPECT diagnosis. Artifical Intelligence in Medicine, 149–169 (2001)Google Scholar
- 4.Bakırcı, Ü., Yıldırım, T.: Diagnosis of Cardiac Problems From SPECT Images by Feedforward Networks, SİU 2004. In: IEEE 12. Sinyal İşleme ve İletişim Uygulamaları Kurultayı, pp. 103–105. Kuşadası (2004)Google Scholar
- 5.Kurgan, L., Cios, K.J.: Ensemble of Classifiers to Improve Accuracy of the CLIP4 Machine Learning Algorithm. In: SPIE’s International Symposium on Sensor Fusion: Architectures, Algorithms, and Applications VI (accepted, 2002)Google Scholar
- 6.UCI Machine Learning Repository, last arrived (February 2007), http://www.ics.uci.edu/~mlearn/MLRepository.html
- 9.Hyvärinen, A., Särelä, J., Vigário, R.: Spikes and bumps: Artefacts generated by independent component analysis with insufficient sample size. In: Int. Workshop on Independent Component Analysis and Blind Separation of Signals (ICA’99), Aussois, France (1999)Google Scholar
- 11.Watkins, A.B.: Exploiting Immunological Metaphors in the Development of Serial, Parallel, and Distributed Learning Algorithms. PhD dissertation, University of Kent, Canterbury (March 2005)Google Scholar