Data, Models, and Scripts of Integrated Physiological Model of Virtual Diabetic Patient for Machine Learning Research

Machine Learning (ML) based predictive models of blood glucose forecasting for diabetic patient produce future glucose trend without any physiological explanation. In this research effort, operation research (OR) is adopted by building a constraint-based physiological model of a diabetic patient with exercise dynamics for producing explanation from ML-based forecasting. The proposed constraint-based diabetic model is implemented in MATLAB/Simulink environment and validated with clinical and free-living datasets. Then the model is applied for OR on forecasting of a Feed-Forward Neural Network (FFNN) which is built on continuous glucose monitoring (CGM) profile along with carbohydrate and physical activity data of a type-2 diabetic patient. This material set contains all Simulink model files, MATLAB scripts, the experimental datasets used for building and validating physiological and ML models, and result files.

基于机器学习（Machine Learning, ML）的糖尿病患者血糖预测模型仅可输出未来血糖变化趋势，却无法提供任何生理学层面的解释。本研究引入运筹学（Operation Research, OR）方法，构建融合运动动力学的糖尿病患者约束生理学模型，以从基于机器学习的预测结果中生成可解释性依据。所提出的约束型糖尿病模型已在MATLAB/Simulink环境中实现，并采用临床数据集与自由生活数据集完成验证。随后将该模型应用于前馈神经网络（Feed-Forward Neural Network, FFNN）预测结果的运筹学分析，该前馈神经网络基于2型糖尿病患者的连续血糖监测（Continuous Glucose Monitoring, CGM）数据、碳水化合物摄入数据与身体活动数据构建。本数据集包含所有Simulink模型文件、MATLAB脚本、用于构建与验证生理学模型及机器学习模型的实验数据集，以及各类结果文件。