Wrapper feature selection with partially labeled data

In this paper, we propose a new feature selection approach with partially labeled training examples in the multi-class classification setting. It is based on modification of genetic algorithm that creates and evaluates candidate subsets during an evolutionary process, taking into account weights recursively eliminating irrelevant features. To increase variety data, unlabeled observations are employed namely by pseudo-labeling them using self-learning recently proposed transductive policy. Empirical results different data sets show effectiveness our method compared to several state-of-the-art semi-supervised approaches.