Motivation: Simulated blood glucose (BG) data is a powerful tool for research, enabling the benchmarking of BG forecasting and control algorithms. However, knowledge-based models provide an unrealistic view of real-world performance, as they lack the features that make real data challenging. At the same time, black-box approaches such as GANs do not enable systematic tests to diagnose model performance.
To address this, we introduce a modular hybrid approach (like knowledge-based methods) and realistic (like data-driven methods) by augmenting simulated data with real data properties. This allows researchers to test algorithms with simulated BG that provide realistic estimates of performance and better understand how different data features contribute to performance on ML tasks.
Louis Gomez, Aishat Toye, R. Stanley Hum, Samantha Kleinberg. Journal of Diabetes Technology and Science, 2023
Paper Link: https://journals.sagepub.com/doi/full/10.1177/19322968231181138