2005 | OriginalPaper | Chapter
An Ensemble of Case-Based Classifiers for High-Dimensional Biological Domains
Authors : Niloofar Arshadi, Igor Jurisica
Published in: Case-Based Reasoning Research and Development
Publisher: Springer Berlin Heidelberg
Activate our intelligent search to find suitable subject content or patents.
Select sections of text to find matching patents with Artificial Intelligence. powered by
Select sections of text to find additional relevant content using AI-assisted search. powered by
It has been shown that an ensemble of classifiers increases the accuracy compared to the member classifiers provided they are diverse. One way to produce this diversity is to base the classifiers on different case-bases. In this paper, we propose the mixture of experts for case-based reasoning (MOE4CBR), where clustering techniques are applied to cluster the case-base into
k
groups, and each cluster is used as a case-base for our
k
CBR classifiers. To further improve the prediction accuracy, each CBR classifier applies feature selection techniques to select a subset of features. Therefore, depending on the cases of each case-base, we would have different subsets of features for member classifiers.
Our proposed method is applicable to any CBR system; however, in this paper, we demonstrate the improvement achieved by applying the method to a computational framework of a CBR system called
TA
3. We evaluated the system on two publicly available data sets on mass-to-charge intensities for two ovarian data sets with different number of clusters. The highest classification accuracy is achieved with three and two clusters for the ovarian data set 8-7-02 and data set 4-3-02, respectively. The proposed ensemble method improves the classification accuracy of
TA
3 from 90% to 99.2% on the ovarian data set 8-7-02, and from 79.2% to 95.4% on the ovarian data set 4-3-02. We also evaluate how individual components in MOE4CBR contribute to accuracy improvement, and we show that feature selection is the most important component followed by the ensemble of classifiers and clustering.