Increasing classification accuracy by combining adaptive sampling and convex pseudo-data

The availability of microarray data has enabled several studies on the application of aggregated classifiers for molecular classification. We present a combination of classifier aggregating and adaptive sampling techniques capable of increasing prediction accuracy of tumor samples for multiclass datasets. Our aggregated classifier method is capable of improving the classification accuracy of predictor sets obtained from our maximal-antiredundancy-based feature selection technique. On the Global Cancer Map (GCM) dataset, an improvement over the highest accuracy reported has been achieved by the joint application of our feature selection technique and the modified aggregated classifier method.