Glycemic-aware metrics and oversampling techniques for predicting blood glucose levels using machine learning

• Published in: PloS one Vol. 14; no. 12; p. e0225613
• Main Authors: Mayo, Michael, Chepulis, Lynne, Paul, Ryan G.
• Format: Journal Article
• Language: English
• Published: United States Public Library of Science 02.12.2019; Public Library of Science (PLoS)
• Subjects: Accuracy; Artificial intelligence; Benchmarking; Biology and Life Sciences; Blood; Blood glucose; Blood Glucose - analysis; Blood Glucose Self-Monitoring - instrumentation; Blood Glucose Self-Monitoring - methods; Business performance management; Computer and Information Sciences; Datasets; Datasets as Topic; Diabetes; Diabetes mellitus; Diabetes mellitus (insulin dependent); Diabetes Mellitus, Type 1 - blood; Diabetes Mellitus, Type 1 - drug therapy; Engineering and Technology; Forecasting; Generalized linear models; Glucose; Humans; Hyperglycemia; Hypoglycemia; Hypoglycemia - blood; Hypoglycemia - chemically induced; Hypoglycemia - diagnosis; Hypoglycemic Agents - administration & dosage; Hypoglycemic Agents - adverse effects; Insulin; Insulin - administration & dosage; Insulin - adverse effects; Knowledge discovery; Laboratory Critical Values; Learning algorithms; Machine learning; Medical research; Medicine and Health Sciences; Methods; Model accuracy; Multilayer perceptrons; Oversampling; Pancreas; Pancreas, Artificial; Patients; Performance measurement; Physical Sciences; Polynomials; Predictions; Preprocessing; Regression models; Research and Analysis Methods; Self Medication - adverse effects; Support Vector Machine; Support vector machines; Technology; Type 1 diabetes; New Zealand; United States > US; Ohio
• ISSN: 1932-6203; 1932-6203
• DOI: 10.1371/journal.pone.0225613

Techniques using machine learning for short term blood glucose level prediction in patients with Type 1 Diabetes are investigated. This problem is significant for the development of effective artificial pancreas technology so accurate alerts (e.g. hypoglycemia alarms) and other forecasts can be generated. It is shown that two factors must be considered when selecting the best machine learning technique for blood glucose level regression: (i) the regression model performance metrics being used to select the model, and (ii) the preprocessing techniques required to account for the imbalanced time spent by patients in different portions of the glycemic range. Using standard benchmark data, it is demonstrated that different regression model/preprocessing technique combinations exhibit different accuracies depending on the glycemic subrange under consideration. Therefore technique selection depends on the type of alert required. Specific findings are that a linear Support Vector Regression-based model, trained with normal as well as polynomial features, is best for blood glucose level forecasting in the normal and hyperglycemic ranges while a Multilayer Perceptron trained on oversampled data is ideal for predictions in the hypoglycemic range.