A novel approach to minimize classifier computational overheads in Big Data using neural networks

摘要

Generally, Big Data denotes that datasets may not be apparent, accomplished, deal with or even acquired by the traditional tools in IT and the software or hardware tools in an acceptable period. This paper presents a new method for the minimization of the For the purpose of feature extraction, the Term Frequency-Inverse Document Frequency (TF-IDF) has been employed. A Feature Selection process takes place using correlation based feature selection (CFS) technique to advance the performance of prediction and further provide some faster predictors that are cost-effective to give a good insight of the fundamental process. Finally, classification process takes place by Particle Swarm optimization (PSO) based support vector machine (SVM) model. As neural networks can create some complex error surfaces using numerous local minima, a Back Propagation Neural Network (BPNN) may also fall into the local minima as opposed to a global minimum. SVMs will further minimize the upper bound of a true error. The PSO algorithm is simple in its implementation, easy to understand and is capable of solving several problems in optimization. The proposed model achieves a minimal misclassification rate of 0.133, precision of 0.846, recall of 0.87 and F-measure of 0.855. (C) 2020 Elsevier B.V. All rights reserved.