Fuzzy Inference Model for Type 2 Diabetes Management: a Tool for Regimen Alterations

Abstract

This paper aims to demonstrate the utility of fuzzy set theory in the design process of a diabetes management system that enables patients to make short term alterations (particularly lifestyle) to their overall regimen as required. The model is a Mamdani Fuzzy Inference System (FIS) configured through domain specific information from experts and recognised diabetes management algorithms. The FIS takes a multi-input multi-output (MIMO) design approach with seven inputs variables (age, gender, weight, height, blood glucose (BG), exercise and diet) and three outputs (glycatedhaemoglobin (A1c), exercise and diet level assessments). Goodness of fit test was conducted based on Mean Square Error (MSE), Normalised Mean Square Error (NMSE) and Normalised Root Mean Square Error (NRMSE) between observed/advised and predicted output values. Overall MSE of 0.0899 shows good fit. For each of the output pairs (A1c, exercise and diet), NRMSE (0.7387, 0.7881 and 0.3716) and NMSE (0.9317, 0.9551 and 0.6051) shows good fit for A1c and exercise, but poor fit for diet. Intelligent models of this sort can help simplify management information for diabetes patients, reduce routine workload for clinicians and allow them to focus more on critical issues. Fully developed, this system can be used to build a database of diabetes management cases that includes daily life event information, ultimately leading to automated care for patients through technology.