OFS: Online Feature Selection based on Correlation and Clustering method along with its application

Abstract

eature Selection is one of the important techniques in Data mining. At the time of dimension reduction for reducing the computational cost and also reduction of noises to improve the classification accuracy the feature selection is an essential technique. In this, its result shows, that almost all learning of feature selection has been restricted to batch learning. Dissimilar to existing batch learning methods, online learning has been chosen by an encouraging familiar of well-organized and scalable machine learning algorithms for large-scale approaches such existing technique is not always appropriate and useful for real-world applications when data instances are of high dimensionality or very expensive to acquire the full set of attributes/features. To highlight this limitation, here found the problem of Online Feature Selection (OFS). The large-scale dataset quantity of online learning needs to retrieve all the features/attributes of occurrence. In OFS it is hard for the online learner to keep a classifier that involves a minimum and fixed or exact number of features. The major challenge of OFS is how to make exact predictions for an iteration using a small number of active features. This article shows two distinct tasks of OFS. The first one is learning with full input in this a learner is allowed to access all the features to decide the subset of active features, and the second is learning with partial input in this only a limited number of features is allowed to be accessed for each iteration by the learner. We have used the Differential Evolutionary (DE) algorithm in this study. The proposed system represents novel techniques such as Multiclass classification, Correlation, and clustering methods to clear up each of the problems and give their performance analysis.