Hybrid Model for Early Diabetes Diagnosis

Abstract

Diabetes Mellitus (silent killer or sugar disease) is a metabolic disease characterized by high glucose levels, either in a body with insufficient insulin to breakdown glucose, or body that is resistant to effects of insulin. To improve early diagnosis, data-mining tools are used to help physicians effectively classify the disease. Study presents a hybrid fuzzy, genetic algorithm trained neural network model as a decision support system for diabetes classification. Adopted data is split into: training, cross validation and testing to aid model validation with appropriate weights and biases set for each variables. Results indicate that age, obesity and family relations (in first and second degree), environmental conditions are critical factors to be watched, While in gestational diabetes, mothers with or without a previous case of GDM is confirmed if there is: (a) history of babies with weight > 4.5kg at birth, (b) resistant to insulin showing polycystic ovary syndrome, and (c) have abnormal tolerance to insulin.