Data Augmentation Methods and their Effects on Long-Range Dependence

Abstract

Data augmentation is a common method for expanding datasets to train machine learning models. In this paper, five different methods are used to obtain augmented sets. In addition, eight measures are used for experimental evaluation of datasets before and after data augmentation methods. The key requirement is that any data augmentation should not alter the fundamental properties and characteristics of the original dataset. This research shows how some data augmentation methods can destroy the long-range dependence of the Internet traffic data (ITD) with distributed denial of service (DDoS) attacks (DDoS ITD), and consequently alter the probability mass function (PMF) and data labelling (tagging) of the DDoS ITD.