Smoking Cessation Recruitment Analysis: A Case Study
Wei Li, Xiaohui Cui, Kevin Michael Amaral, Rajani Shankar Sadasivam, Ping Chen · 2017
The primary goal of this paper is to generate simulated data which is useful when only limited data is available. We introduce two techniques in this study to augment the datasets: (1) change a very small set of fields' values randomly and (2) through using generative adversarial networks (GANs). We propose a few analysis methods on classification problems to improve the accuracy of a well-sought class: (1) remove border samples between two categories, (2) reduce dimensionality through feature selection, (3) sacrifice the accuracy of less-valuable classes. We applied these methods to a real-word dataset: Smoking Cessation groups. One of the biggest challenges in this vein is that there is little available data, which is often the case in medical fields where data collection can be expensive and difficult. Also, a small amount of data may not contain sufficient information for machine learning methods to generate generalizable results. There are many existing methods to deal with this problem. However, their performance needs to be significantly improved in practice. Our results show that applying each of these analysis methods improves classification accuracy of the well-sought class and proved the GANs can generate many simulation data.