An online approach for feature selection for classification in big data

Abstract

Feature selection (FS), also known as attribute selection, is a process of selection of a subset of relevant features used in model construction. This process or method improves the classification accuracy by removing irrelevant and noisy features. FS is implemented using either batch learning or online learning. Currently, the FS methods are executed in batch learning. Nevertheless, these techniques take longer execution time and require larger storage space to process the entire dataset. Due to the lack of scalability, the batch learning process cannot be used for large data. In the present study, a scalable efficient Online Feature Selection (OFS) approach using the Sparse Gradient (SGr) technique was proposed to select the features from the dataset online. In this approach, the feature weights are proportionally decremented based on the threshold value, which results in attaining zeros for the insignificant features' weights. In order to demonstrate the efficiency of this approach, an extensive set of experiments was conducted using 13 real-world datasets that range from small to large size. The results of the experiments showed an improved classification accuracy of 15%, which is considered to be significant when compared with the existing methods.

Online Feature Selection (OFS) with Accelerated Bat Algorithm (ABA) and Ensemble Incremental Deep Multiple Layer Perceptron (EIDMLP) for big data streams

“…The assessment and implementation uncovered the competency of these algorithms and further leads to new research direction in FS problems. Nazar and Senthilkumar [21] contributed an efficient, scalable OFS, which used the Sparse Gradient (SGr) for the online selection of features. In this approach, based on the threshold value, the feature weights were proportionally decremented, which zeroed irrelevant featured weights.…”

“…OFS [14] is related to streaming features. The interpretation of OFS [21] is represented by the notation Ds = [Ds 1 , Ds 2 ,…,Ds n ] T ∈ R n×d, Where Ds 1 ,…,Ds n is the given dataset with the feature set Fs = [fs 1 , fs 2 ,…, fs d ] T ∈ R d and let Cl = [cl 1 , cl 2 ,…,cl m ] T ∈ R m denote the class label vector. Let d be the number of features which is unknown in priori, the best (1) n ′ = n − min ds max ds -min ds (nfmax −nfmin) + nfmin feature subsets are selected from d such that s < d. Accuracy will be achieved through only selecting the most relevant feature subset for classification.…”

“…Datasets characteristics Confusion matrixF-measure is the ratio precision and recall, given in Eq (21)…”

A novel hybrid teaching-learning-based optimization algorithm for the classification of data by using extreme learning machines

“…Data classification is a big challenge for scientists when they need to extract any useful knowledge contained by the data and answer some important questions related to the patterns of data [1,2]. Many data-mining and machine learning techniques have been proposed for the solution of this important problem.…”

Determining Weighted, Utility-Based Time Variant Association Rules Using Frequent Pattern Tree

“…Utilitybased frequency-weighted mining problems are used to locate frequent item-sets which have higher utility than a user specified minimum-one. Unlike a priori-algorithm [2], [3], which is affirmed on support and confidence framework, mining of frequency-weighted recurrent item-sets on their utility-base from time-variant transactional dataset poses a greater challenge [4], [5].…”