Handling Label Noise in Microarray Classification with One-Class Classifier Ensemble

Abstract

The advance of high-throughput techniques, such as gene microarrays and protein chips have a major impact on contemporary biology and medicine. Due to the high-dimensionality and complexity of the data, it is impossible to analyze it manually. Therefore machine learning techniques play an important role in dealing with such data. In this paper, we investigate the influence of label noise on the effectiveness of classification system applied to microarray analysis. Popular methods do not have any mechanism for handling such difficulties embedded in the nature of data. To cope with that, we propose to use a one-class classifiers, which distinct from canonical methods, rely on objects coming from single class distributions only. They distinguish observations coming from the given class from any other possible decision about the examples, that were unseen during the classification step. While having less information to dichotomize between classes, one-class models can easily learn the specific properties of a given data set and are robust to difficulties embedded in the nature of the data. We show, that using ensembles of one-class classifiers can give as good results as canonical multi-class classifiers, while allowing to deal with unexpected label noise in the data. Experimental investigations, carried out on public data sets, prove the usefulness of the proposed approach.