Researchers from the Faculties of Science and Medicine Develop CHDS to Enhance AI and Health Data Security

25 September 2026
Corporate Communication and Alumni Relations Center (CCARC)
Researchers from the Faculties of Science and Medicine, Chiang Mai University, have collaborated to develop CHDS: Geometry-based Synthetic Data Generation for Tabular Data, a method designed to overcome privacy and data accessibility limitations while advancing artificial intelligence (AI) development and data analytics. As AI and data analytics play increasingly important roles across various fields, access to real-world data remains challenging due to privacy concerns, data sensitivity and data insufficiency, particularly in medicine and public health. To address these limitations, the researchers developed CHDS as an alternative approach to synthetic data generation. The method generates synthetic data that preserves the key characteristics of real-world data without disclosing personal information. A key advantage of CHDS is its geometry-based approach to synthetic data generation for tabular data, which does not require assumptions about a specific data distribution pattern. The proposed method incorporates a convex hull to define the boundaries of the data, density estimation to estimate data distributions, and Markov Chain Monte Carlo (MCMC)-based heuristic random sampling to generate new data samples. It can efficiently accommodate datasets containing both numerical and categorical variables. The method achieves a balance between data utility and privacy, as demonstrated through a comparative evaluation of six state-of-the-art synthetic data generation methods across four machine learning datasets. CHDS is particularly useful in situations where access to real-world data is restricted, such as when developing and testing machine learning models without fully disclosing or publishing sensitive data. From an academic perspective, this research introduces a novel approach to synthetic data generation by integrating geometry-based concepts, density estimation and MCMC, offering an alternative to conventional methods such as generative adversarial networks (GANs), variational autoencoders (VAEs) and diffusion-based methods. In terms of social benefits, the method can facilitate the use of data in AI research and development while reducing the need to share sensitive real-world data. Economically, it can help alleviate data accessibility constraints in the development and testing of data analytics and machine learning systems across various sectors. The research was published in Knowledge-Based Systems (Volume 351, Part B, Article 116745) in 2026.

The full article is available at https://doi.org/10.1016/j.knosys.2026.116745.

The research team comprised Mr. Sopanat Phongmaruthai, a master's student in Computer Science; Assistant Professor Dr. Suphanat Chaidee, Department of Mathematics; Dr. Chanthanok Aramrat and Associate Professor Dr. Chaisiri Angkurawaranon, Department of Family Medicine, Faculty of Medicine; and Assistant Professor Dr. Paphangkorn Inkaew, Department of Computer Science, Faculty of Science (Corresponding Author).

Gallery