The Generation of Synthetic Healthcare Data Using Deep Neural Networks
Year: 2022
Type: dissertation
Abstract: High-quality tabular data is a crucial requirement for developing data-driven applications,\nespecially healthcare-related ones, because most of the data nowadays\ncollected in this context is in tabular form. However, strict data protection laws introduced\nin Health Insurance Portability and Accountability (HIPAA) and General\nData Protection Regulation (GDPR) present many obstacles to accessing and doing\nscientific research on healthcare datasets to protect patients’ privacy and confidentiality.\nThus, synthetic data has become an ideal alternative for data scientists and\nhealthcare professionals to circumvent such hurdles. Although many healthcare data\nproviders still use the classical de-identification and anonymization techniques for\ngenerating synthetic data, deep learning-based generative models such as Generative\nAdversarial Networks (GANs) have shown a remarkable performance in generating\ntabular datasets with complex structures. Thus, this thesis examines the GANs’\npotential and applicability within the healthcare industry, which often faces serious\nchallenges with insufficient training data and patient records sensitivity.\n\nWe investigate several state-of-the-art GAN-based models proposed for tabular synthetic\ndata generation. Precisely, we assess the performance of TGAN, CTGAN,\nCTABGAN and WGAN-GP models on healthcare datasets with different sizes,\nnumbers of variables, column data types, feature distributions, and inter-variable\ncorrelations. Moreover, a comprehensive evaluation framework is defined to evaluate\nthe quality of the synthetic records and the viability of each model in preserving\nthe patients’ privacy. After training the selected models and generating\nsynthetic datasets, we evaluate the strengths and weaknesses of each model based\non the statistical similarity metrics, machine learning-based evaluation scores, and\ndistance-based privacy metrics.\n\nThe results indicate that the proposed models can generate datasets that maintain\nthe statistical characteristics, model compatibility, and privacy of the original ones.\nMoreover, synthetic tabular healthcare datasets can be a viable option in many\ndata-driven applications. However, there is still room for further improvements in\ndesigning a perfect architecture for generating synthetic tabular data.
Author: Hossein Zadeh Nik, Alireza
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6991382996