Forward Semi-supervised Feature Selection Based on Relevant Set Correlation

Feature selection is among the keys in many applications, especially in mining high-dimensional data. With lack of labeled instances, the learning accuracy may deteriorate using traditional methods. In this paper, we introduce a ldquowrapperrdquo type semi-supervised feature selection approach based on RSC model. It extends the class label from labeled training set to unlabeled data. Additionally, we consider the case of overlapping during the extension. With respect to the experiments, our algorithm is proved to have a promising performance on the improvement of learning accuracy.