Support Vector Regression 기반 개인화 혈당 예측 알고리즘

Abstract

Personalized glucose prediction algorithm (PGPA) is considered as an excellent approach to manage glucose levels due to abilities to consider a patient‘s non-linear glucose patterns. To extract continuous glucose monitoring (CGM) time-series data, 30 virtual patients with type 1 diabetes were generated by UVA/Padova T1DMS. The developed support vector regression that was trained with CGM points collected for 3 days showed 17.7 mg/dL of root mean square errors and 11.6 % of mean absolute percentage error on average. In conclusion, we validated the approach of PGPA with the patients and it will be greatly helpful to manage blood glucose level.