Feature Selection for Improving Case-Based Classifiers on High-Dimensional Data Sets

Case-based reasoning (CBR) is a suitable paradigm for class discovery in molecular biology, where the rules that define the domain knowledge are difficult to obtain, and there is not sufficient knowledge for formal knowledge representation. To extend the capabilities of this paradigm, we propose logistic regression for CBR (LR4CBR), a method that uses logistic regression as a feature selection (FS) method for CBR systems. Our method not only improves the prediction accuracy of CBR classifiers in biomedical domains, but also selects a subset of features that have meaningful relationships with their class labels. In this paper, we introduce two methods to rank features for logistic regression. We show that using logistic regression as a filter FS method outperforms other FS techniques, such as Fisher and t-test, which have been widely used in analyzing biological data sets. The FS methods are combined with a computational framework for a CBR system called TA3 . We also evaluate the method on two mass spectrometry data sets, and show that the prediction accuracy of TA3 improves from 90% to 98% and from 79.2% to 95.4%. Finally, we compare our list of discovered biomarkers with the lists of selected biomarkers from other studies for the mass spectrometry data sets, and show the overlapping biomarkers.