The increased interest in data privacy-conscientious machine-learning processes has promoted the use of synthetic data as an effective substitute to real data. The most popular frameworks in terms of the production of high-fidelity synthetic data have become Generative Adversarial Networks (GANs) because they are capable of capturing complex and high-dimensional distributions. The paper introduces an overall synthetic data generation approach based on adversarial training, including a deep GAN architecture, which is based on the WGAN variant of the architecture, namely, WGAN-GP variant. The quality of the generated data is evaluated by a multi-dimensional evaluation framework which includes statistical similarity, utility, and privacy. The experimental outcomes indicate that synthetic datasets created with the help of GAN can reach a high level of similarity to actual data and minimize the risk of privacy considerably. The paper ends with a set of recommendations on how synthetic data pipelines can be deployed in environments where privacy is at stake.