Data mining for the diagnosis of type II diabetes from three-dimensional body surface anthropometrical scanning data

Diabetes mellitus has become a general chronic disease as a result of changes in customary diets. Impaired fasting glucose (IFG) and fasting plasma glucose (FPG) levels are two of the indices which physicians use to diagnose diabetes mellitus. Although this is a fairly accurate approach, the tests are expensive and time consuming. This study attempts to construct a prediction model for Type II diabetes using anthropometrical body surface scanning data. Four data mining approaches, including backpropagation neural network, decision tree, logistic regression, and rough set, were used to select the relevant features from the data to predict diabetes. Accuracy of classification was evaluated for these approaches. The result showed that volume of trunk, left thigh circumference, right thigh circumference, waist circumference, volume of right leg, and subjects' age were associated with the condition of diabetes. The accuracy of the classification of decision tree and rough set was found to be superior to that of logistic regression and backpropagation neural network. Several rules were then extracted based on the anthropometrical data using decision tree. The result of implementing this method is not only useful for the physician as a tool for diagnosing diabetes, but it is sophisticated enough to be used in the practice of preventive medicine.