Comparison of three data mining models for predicting diabetes or prediabetes by risk factors

Abstract

The purpose of this study was to compare the performance of logistic regression, artificial neural networks (ANNs) and decision tree models for predicting diabetes or prediabetes using common risk factors. Participants came from two communities in Guangzhou, China; 735 patients confirmed to have diabetes or prediabetes and 752 normal controls were recruited. A standard questionnaire was administered to obtain information on demographic characteristics, family diabetes history, anthropometric measurements and lifestyle risk factors. Then we developed three predictive models using 12 input variables and one output variable from the questionnaire information; we evaluated the three models in terms of their accuracy, sensitivity and specificity. The logistic regression model achieved a classification accuracy of 76.13% with a sensitivity of 79.59% and a specificity of 72.74%. The ANN model reached a classification accuracy of 73.23% with a sensitivity of 82.18% and a specificity of 64.49%; and the decision tree (C5.0) achieved a classification accuracy of 77.87% with a sensitivity of 80.68% and specificity of 75.13%. The decision tree model (C5.0) had the best classification accuracy, followed by the logistic regression model, and the ANN gave the lowest accuracy.

Keywords

• Community
• Data mining
• Diabetes
• Prediction
• Risk factors