An efficient machine-learning model based on data augmentation for pain intensity recognition

doi:10.1016/j.eij.2020.02.006

2020-12-01

Ahmad Al–Qerem *

DOI：10.1016/j.eij.2020.02.006

原文链接

查看原文

摘要

En 中文

Pain is defined as a distressing experience associated with actual or potential tissue damage with sensory, emotional, cognitive and social components, knowing the exact level of pain experienced to have a critical impact for caregivers to make diagnosis and make he suitable treatment plan, but the available methods depend entirely on the patient self-report, which increase the difficulties of knowing the accurate level of pain experienced by the patient. Therefore, automating this process became an important issue, but due to the hardness of acquiring medical data, it became difficult to build a predictive model with good performance. Generative Adversarial Networks is a framework that generates artificial data with a distribution similar to the real data, by training two networks; the generator which tries to generate new samples similar to the real ones, and the discriminator which applies a traditional supervised classification to distinguish the augmented samples, the optimal case is when the discriminator cannot distinguish the augmented samples from the real samples. In this research, we generated data using Least Square Generative Adversarial Networks and the study the effect of applying feature selection on the data before the augmentation. Moreover, the approach was tested on a dataset that contains multi biopotential signals for different levels of pain. (C) 2020 Production and hosting by Elsevier B.V. on behalf of Faculty of Computers and Artificial Intelligence, Cairo University. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Keyword:

Machine-learning

Data augmentation

Pain intensity recognition

Features selection

GANs