| dc.description.abstract | In this work, we present a novel algorithm that considers attributes from different experimental sources as separate groups for the purpose of classification. We remove noise from each of these groups, combine them, and then run the classifier on the grouped data. Examples are considered to be noise if they do not contribute to the prediction but, rather, degrade the quality of the classification result. As part of this work, we identify a measure that appropriately labels noise without knowledge of the class labels. Our method shows that the classification result is better when run on such filtered, grouped data than when run on the entire grouped data. In this work, we have considered time-series data because of their noisy nature. Our approach can be viewed as unsupervised feature-subset selection in grouped attributes and at the level of each instance individually. | en_US |
