A Comparative Study of Recent Feature Selection Techniques Used in Text Classification

Abstract

As we all know, handling large amounts of data is a problem these days. Despite having so many resources to store, train and process the data, still it is required to reduce these datasets in order to reduce computational complexity, save time, cost and retrieve valuable information from large text documents. The presentation of a machine learning algorithm relies upon the dataset utilized. When the dataset is large, the learning algorithm tries to accommodate all the features which increases the dimensionality of the data. This high-dimensional data is not useful as it might contain irrelevant and redundant features. It becomes important to remove these features. Thus, pre-processing of data is required to compress and analyse the dataset for the purpose of text classification (TC). This can be achieved by using feature selection (FS) techniques. The fundamental goal of FS techniques is to acknowledge pertinent features and to get rid of repetitive attributes w.r.t. high-dimensional data (Shroff and Maheta in 2015 International Conference on Computer Communication and Informatics (ICCCI), Coimbatore, India, 2015 [1]). Nowadays, major FS methods use optimization algorithms (Brownlee in https://machinelearningmastery.com/. 23 Dec 20 [Online]. Available: https://machinelearningmastery.com/tour-of-optimization-algorithms/. Accessed 3 Feb 2021 [2]) to get an ideal component subset from high-dimensional information from feature space which decreases computational expense and builds classifier precision. Some of the recent feature selection techniques have been discussed in this paper which can prove to be useful for text classification (TC).