A novel semi-feature selection method based on hybrid feature selection mechanism
Shangzhi Zheng, Hualong Bu · 2010
Many Semi-supervised learning applications require a feature selection method to deal with the unlabeled samples. Traditional researches deal it either with the “filter-type” feature selection mechanism, which may not work well for classification tasks or “wrapper” mechanism, which need high computational cost. Here we proposed a new semi-supervised feature selection method based on hybrid feature selection mechanism. Its principle lies in using Relief Wrapper method to explore the usage of unlabeled examples, which will help for training classifiers. In essence, it uses unlabeled examples to extend the initial labeled training set with the help of classifiers. Extensive experiments on publicly available datasets and formal analysis show its nice combination of efficiency and accuracy