Publications (1)0 Total impact
- SourceAvailable from: Dirk Aeyels[Show abstract] [Hide abstract]
ABSTRACT: To summarize, this work studied the effect of feature selection techniques for classifying nucleic acid sequences. The scientific contribution of this study comprises several aspects. The first aspect concerns the algorithmical part of the work, where we made advances in the field of pattern recognition, developing new methods for feature selection, feature ranking and feature weighting. A second aspect of the work is the application of these techniques to the task of gene prediction. For a chosen subset of classification problems within the gene prediction task, we showed how feature selection techniques can be used to improve classification performance and to extract new knowledge about the biological processes we are modelling. Furthermore, we obtained some more general insights in the application of feature selection techniques. We showed that different techniques might select different features, and that care has to be taken when interpreting the results obtained by feature selection. A robust way of obtaining reliable domain knowledge is to compare different feature selection techniques and different classifiers, and to identify common patterns in the features that are selected as relevant.