Bagesh Kumar*, Akhil Shukla, Akhil Singh, Mohd Javed Ali and OP Vyas
IIIT Allahabad, India
*Corresponding Author: Bagesh Kumar, IIIT Allahabad, India.
Received: February 07, 2022; Published: May 25, 2022
The objective of this paper is to minimize the number of samples required for training algorithms involving support vectors while maximizing knowledge of the target class. A method is proposed which uses autoencoder in conjunction with farthest boundary point extraction for selecting most promising frontier points from the original sample. Farthest frontier points are chosen using a geometrical approach for estimating extreme points of a class and autoencoder for learning a compressed representation of the data. For experimentation, we have used datasets of MNIST, Iris, credit card fraud detection, Indian Pines, Human Activity Recognition Database.
Keywords: Sample Reduction; Autoencoder; Dimensionality Reduction; Farthest Boundary Point Extraction; Multiclass Classification; SVM; Training Data Reduction
Citation: Bagesh Kumar., et al. “Reduction of Training Data from Large Datasets using Encoder and Decoder Algorithm without Loss of Accuracy". Acta Scientific Computer Sciences 4.6 (2022): 59-74.
Copyright: © 2022 Bagesh Kumar., et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.