|
Abstract:
|
Self-reported continuous Glucose Monitor Sensor data (CGM) and insulin pump data are analyzed by health care providers to learn about and optimize the health of diabetic individuals. CGM is an example of the volume, velocity and variety of time-stamped data brought about by the IoT. Such volume is not being met by an increasing interest of Colleges in the teaching of time series to undergraduates. The “Introduction to Time Series” course for undergraduates, if offered at all (which is rarely the case), is an elective taken by a small number of students who are exposed for the first time to the requirement of taking into account the non i.i.d. nature of the data. The case study presented in this paper offers an opportunity for teachers of introductory statistics to motivate students to learn "features" of time series data such as: autocorrelation, seasonality, trends, decomposition, modeling, prediction and prediction error. The “forecast” of blood glucose level for individual patients is a very active area of machine learning research for time series data in health applications, and offers an opportunity to satisfy student’s increasing interest in machine learning as well.
|
