Methodological findings from generating multidimensional synthetic datasets

This paper provides an overview of key approaches for creating multidimensional synthetic datasets, including probabilistic concatenation, scaling with business rules, inverse copula sampling, and generative adversarial networks. These methods address challenges such as data scarcity, privacy preservation, and maintaining representational accuracy. A methodological tree is introduced to systematize cases where synthetic data generation is essential, from handling incomplete datasets to enriching existing ones. Emphasis is placed on preliminary audits for data minimization, ensuring computational efficiency and enhancing validation feasibility. While this paper is not empirical, it synthesizes theoretical insights and offers a synthesized dataset for reference. The findings contribute to establishing best practices in synthetic data generation, enabling researchers to meet the demands of modern data-driven studies effectively.