Predictive glucose concentration modeling in type 1 diabetes patients by application of Reservoir Computing

Bankosegger, R. (2018) Predictive glucose concentration modeling in type 1 diabetes patients by application of Reservoir Computing. Bachelor's Thesis, Artificial Intelligence.

Preview	Text Bachelor_AI_research_project_2_1.pdf - Published Version Download (795kB) | Preview
	Text toestemming.pdf - Other Restricted to Backend only Download (78kB)

Abstract

Predictive modeling of glucose concentration has the potential to positively influence Diabetes Mellitus therapy by alerting to hyper- or hypoglycemic events and by providing pa- tients with a reasoning framework for short-term treatment decisions. Literature describes the successful application of data-driven techniques such as feed-forward neural networks and sup- port vector regression. In this study, the viability of Reservoir Computing as a novel approach is investigated. This is motivated by the fact that in some instances of time series modeling Reservoir Computing has yielded more accurate results than the above mentioned techniques. The proposed model, an Echo State Network, is based on subcutaneous blood glucose, carbo- hydrate intake, bolus insulin intake and time of day. Rolling-origin evaluation for forecasting horizons up to 120 minutes was performed on data collected from three patients in free-living conditions. Results show that the Echo State Network consistently outperformed the control model (assuming no change in glucose concentration) by achieving lower error rates. Although these results are promising, the examined model was unable to outperform the control model in terms of clinical accuracy, which was attributed to the network’s inability to predict fast changes in glucose concentration.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Wiering, M. and Netten, S. van
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	15 Feb 2018 08:35
Last Modified:	02 May 2019 09:49
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/16454

Actions (login required)

View Item