Classification of Type 2 Diabetes Using Machine Learning Techniques

Abstract

Diabetes is a lifelong chronic disease defined by disorders in protein, fat and carbohydrate metabolism as a result of complete or partial deficiency of insulin hormone secreted from the pancreas. This disease is caused by the absence or deficiency of insulin hormone in the body. Normal metabolism also breaks down in the intestines to convert nutrients into glucose. Then, when this glucose passes through the intestines into the blood, the level of sugar in the blood rises. In healthy people, glucose in the blood is transported to cells with the help of insulin hormone, which is secreted from the pancreas. Because sugar can not be transported to the cell if there is a deficiency or impaired effect of insulin hormone in the body, glucose increases in the blood and develops an increase in blood sugar (hyperglycemia), called diabetes. Early diagnosis of diseases that will occur in insulin, which is vital for the human body, is of great importance. The aim of this study is to use machine learning techniques to diagnose Type 2 diabetes using medical laboratory data. As machine learning techniques, J48, Random Forest, Random Tree and IBk algorithms in the WEKA programme were used. In this study, 400 patient data were investigated. 6 laboratory tests such as age, gender, glucose, HbA1C, HGB and urine were selected as input data. All four algorithms used were successfully trained. The highest accuracy value was found 96.97% in Random Forest algorithm, with recall and F-measure values of 98.47% and 96.24%, respectively.