A Survey on Online Feature Selection

Feature selection is one of the main techniques used in data mining. In defiance of its consequence, most learning of feature selection is limited to batch learning. Dissimilar to existing batch learning methods, online learning can be elected by an encouraging family of well-organized and scalable machine learning algorithms for large-scale approach. The greatest quantity of online learning need to retrieve all the attributes/features of occurrence. Such a simple surroundings are not invariable for real-world applications when statistics illustration is of high-dimensionality. The problem of Online Feature Selection (OFS) is that online learner is allowed to maintain a classifier which involved only a small and fixed number of features. Online feature selection is to make accurate prediction for an object using a small number of active features. This article addresses two different tasks of online feature selection 1) learning with full input 2) learning with partial input. The proposed system presents novel algorithm to solve each of the two problems and give their performance analysis.