Measuring inter-dataset similarity is an important task in machine learning and data mining with various use cases and applications. Existing methods for measuring inter-dataset similarity are computationally expensive, limited, or sensitive to different entities and non-trivial choices for parameters. They also lack a holistic perspective on the entire dataset. In this paper, we propose two novel metrics for measuring inter-dataset similarity. We discuss the mathematical foundation and the theoretical basis of our proposed metrics. We demonstrate the effectiveness of the proposed metrics by investigating two applications in the evaluation of synthetic data and in the evaluation of feature selection methods. The theoretical and empirical studies conducted in this paper illustrate the effectiveness of the proposed metrics.

Synthetic data is by many expected to have a significant impact on data science by enhancing data privacy, reducing biases in datasets, and enabling the scaling of datasets beyond their original size. However, the current landscape of tabular synthetic data generation is fragmented, with numerous frameworks available, only some of which have integrated evaluation modules. synthesizers is a meta-framework that simplifies the process of generating and evaluating tabular synthetic data. It provides a unified platform that allows users to select generative models and evaluation tools from open-source implementations in the research field and apply them to datasets of any format. The aim of synthesizers is to consolidate the diverse efforts in tabular synthetic data research, making it more accessible to researchers from different sub-domains, including those with less technical expertise such as health researchers. This could foster collaboration and increase the use of synthetic data tools, ultimately leading to more effective research outcomes.